Characterization of first trimester fetal erythroblasts for non‐invasive prenatal diagnosis

Abstract

Isolating fetal erythroblasts from first trimester maternal blood offers a promising non‐invasive alternative for prenatal diagnosis. The aim of this study was to characterize the biological properties of first trimester primitive erythroblasts to facilitate their enrichment from first trimester maternal blood. Primitive erythroblasts were the predominant cell type until 12 weeks gestation, after which time their numbers declined steeply; 100% were ϵ‐globin‐positive versus <0.06% definitive erythroblasts. Buoyant densities of first trimester fetal erythroblasts ranged from 1.077 to 1.130 g/ml, and optimal recoveries were obtained with Percoll 1118. Although primitive erythroblasts carried a negative surface charge and were resistant to NH₄Cl lysis, these properties had only a limited role in fetal cell enrichment. Immunophenotyping showed that primitive, like definitive, erythroblasts were GPA+, CD47+, CD45– and CD35–, whereas CD71 expression was weak/undetectable on primitive erythroblasts but strongly positive on 100% of definitive erythroblasts; primitive erythroblasts were also CD36– whereas definitive erythroblasts were CD36+. We therefore used CD45/GPA selection of Percoll 1118‐separated cells to demonstrate successful enrichment of male ϵ‐globin‐positive fetal erythroblasts from model mixtures, and as proof of principle from some first trimester maternal blood samples. Fetal cell enrichment protocols based on first trimester ϵ‐globin‐positive primitive erythroblasts may allow reliable enrichment of fetal cells from maternal blood for early non‐invasive prenatal diagnosis of genetic disorders.

Submitted on October 21, 2002; resubmitted on December 18, 2002. accepted on December 30, 2002

Introduction

Isolation of fetal nucleated red blood cells (NRBC) from maternal blood would allow non‐invasive prenatal diagnosis of chromosomal and monogenic disorders without the inherent risks of invasive procedures (Bianchi et al., 2002). Components of this technique include enrichment, identification and molecular genetic diagnosis using chromosomal fluorescence in‐situ hybridization (cFISH) or polymerase chain reaction (PCR). With regard to enrichment and diagnosis, we and others (Boye et al., 2001; Choolani et al., 2001; Hogh et al., 2001; Voullaire et al., 2001), have demonstrated that ϵ‐globin is a highly specific marker for first trimester fetal erythroblasts. We have also described a method for combining ϵ‐globin identification with cFISH within the same cell, for diagnosis within cells definitively identified as fetal (Choolani et al., 2001). However, enriching fetal cells from first trimester maternal blood has proved challenging because of their rarity in maternal blood (Bianchi et al., 1997), and because enrichment protocols have been extrapolated from cord blood (Bhat et al., 1990; 1993; Oosterwijk et al., 1998; Smits et al., 2000) and second trimester NRBC (Zheng et al., 1999; Rodriguez De Alba et al., 2001), since first trimester fetal blood is not readily available. This may explain why enrichment of primitive erythroblasts from first trimester maternal blood has never been described. Several potential differences between first and later trimester fetal erythroblasts could be exploited to enrich primitive erythroblasts from maternal blood (size and morphology, surface antigens, surface charge, buoyant density and resistance to lysis).

First trimester fetal blood contains two types of nucleated erythrocytes: primitive erythroblasts and definitive NRBC (Palis and Segel, 1998; Keller et al., 1999). The primitive cells, derived from the yolk sac (Palis and Yoder, 2001), are large and have a high cytoplasmic:nuclear ratio. Early in maturation, their nuclei have a fine reticular appearance that becomes condensed and pyknotic with cell differentiation (Kelemen et al., 1979). In contrast, definitive erythroblasts, produced in the fetal liver and bone marrow, are smaller with a lower cytoplasmic:nuclear ratio and extrude their nuclei on maturation (Keller et al., 1999). The site of erythropoiesis switches from yolk sac to liver by ∼6 weeks (Peschle et al., 1985) but the persistence, rate of fall and ultimate fate of primitive erythroblasts circulating in fetal blood have not been characterized.

No surface antigens specific for first trimester fetal erythroblasts have been identified (Huie et al., 2001) so it is likely that a combination of antibodies will be needed to separate these cells from nucleated cells of maternal origin. Principal antigens expressed on adult and/or cord erythroid cells, and therefore potential identifiers of fetal erythroid cells, are the transferrin receptor (CD71), glycophorin A (GPA), thrombospondin receptor (CD36), complement receptor 1 (CD35) and integrin‐associated protein (CD47, IAP) (Loken et al., 1987; Bianchi et al., 1990; Price et al., 1991; Cheung et al., 1996; Durrant et al., 1996; Oldenborg et al., 2000; Telen et al., 2000; Zhong et al., 2000; Al‐Mufti et al., 2001). CD71 has been widely used to enrich fetal NRBC since it is expressed not only in cord and adult NRBC (Loken et al., 1987; Bianchi et al., 1990; Price et al., 1991; Cheung et al., 1996; Durrant et al., 1996; Zhong et al., 2000; Al‐Mufti et al., 2001), but also in second trimester erythroblasts (Zheng et al., 1999). However, the extent of CD71 expression on first trimester fetal erythroblasts is unknown. Similarly, the expression profile of GPA, CD35, CD36 and CD47 (Oldenborg et al., 2000; Telen et al., 2000) on first trimester erythroid cells has not been characterized.

An alternative approach for enriching fetal cells from maternal blood is to exploit differences in the physical properties of fetal compared with maternal cells, including surface charge, density or susceptibility to lysis. Charge flow separation, which sorts cells according to their characteristic surface charge, proportional to the sialic acid content of their glycocalyx (Suzuki et al., 1998), has been used to enrich negatively charged fetal NRBC although it is unclear what role charge, as opposed to cell size, played in the separation (Wachtel et al., 1996; 1998; Shulman et al., 1998). Although density centrifugation is widely used to enrich fetal from maternal cells, the buoyant density of first trimester fetal erythroblasts is unknown.

Adult erythrocytes are more susceptible to ammonium chloride lysis than fetal erythroblasts (Alter et al., 1979; De Graaf et al., 1999), as carbonic anhydrase activity is ≥5‐fold greater and acetazolamide permeability ∼10‐fold less in adult compared with fetal red blood cells (Boyer et al., 1976). Although it has been assumed that fetal cells remain unchanged during this process, one recent study found a 60‐fold increase in fetal NRBC following ammonium chloride lysis, suggesting that fetal NRBC are altered by lysis (Voullaire et al., 2001).

We characterized the morphological, immunophenotypic and physical characteristics of first trimester fetal erythroblasts, and investigated, using a simple enrichment protocol, how these might be combined to enrich ϵ‐globin‐positive fetal primitive erythroblasts from first trimester maternal blood.

Materials and methods

Blood and tissue collection for research was approved by the institutional ethics committee, in compliance with national guidelines regarding the use of fetal tissue for research purposes. All subjects gave written informed consent.

First trimester fetal erythroblasts

Fetal whole blood (n = 36) was collected by ultrasound‐guided transabdominal cardiocentesis before surgical termination of apparently normal pregnancies for social indications (Campagnoli et al., 2000). Fetal gestational ages (x weeks^{+y days}), as determined by ultrasonically measured crown–rump length, ranged between 7⁺⁰ and 13⁺⁶ weeks. Nucleated cell concentrations within fetal blood were calculated using a haemocytometer and erythroblast frequency was determined by examining 200 cells per slide after Wright’s staining. Relative frequencies of primitive and definitive lineage erythroblasts [the morphological features of each are described in the introduction (Kelemen et al., 1979; Peschle et al., 1985)] within circulating fetal blood were determined by immunostaining all fetal NRBC with anti‐GPA. This ensured that only erythroid nucleated cells were analysed and the uniform staining of both types of fetal erythroblasts allowed the study of their changing proportions in the first trimester.

We previously reported that the percentage of ϵ‐positive fetal erythroblasts declines through the first trimester to negligible levels by 14 weeks (Choolani et al., 2001). To determine the frequency of primitive and definitive NRBC expressing ϵ‐globin, nine representative slides from that dataset (9–12 weeks) stained for ϵ‐globin (Choolani et al., 2001) were selected and 200 cells of each lineage examined.

Surface antigen expression

Surface antigen expression of GPA, CD45 (common leukocyte antigen), CD71 (transferrin receptor) and CD47 (IAP) (BD Biosciences Pharmingen, USA) by primitive and definitive erythroblasts was determined by alkaline phosphatase immunocytochemistry, with only minor modifications to our previously published protocol (Choolani et al., 2001). Adult erythrocytes were used as positive controls for GPA and CD47 and as negative controls for CD45 and CD71. Adult lymphocytes were used as positive controls for CD45 and negative controls for GPA. K562, an erythroleukaemia cell line, was used as a positive control for CD71. Internal negative controls for CD47, GPA, CD45 and CD71 involved omitting primary antibody, secondary antibody or label, or the use of an irrelevant primary antibody (anti‐cytokeratin). Two hundred adjacent cells were examined on each control slide. Fourteen test slides were stained for each antigen and 500 cells scored. CD71 expression was determined in five samples (9⁺⁰ to 11⁺⁶ weeks) using the luminosity histogram function on graphics software (Adobe Photoshop 5.5; Adobe Systems UK). Photographs of GPA‐ and CD71‐stained cells were digitized by reflective scanning at 300 dots per inch. Mean pixel intensity from up to five small squares of positive cells (105–1861 pixels) was determined and luminosity (brightness) determined on 256 grey scale (arbitrary units, AU).

For FITC‐conjugated anti‐CD36 (BD), monocytes were used as positive, and erythrocytes as negative, antigen controls. Internal positive controls were FITC‐labelled anti‐GPA and FITC‐labelled anti‐CD71, and the internal negative control, FITC‐labelled anti‐CD45. Since CD35 is expressed at very low density on erythrocytes and loses its binding potential if exposed to harsh fixatives, anti‐CD35 (BD) was used at a concentration of 1:10, with fixation and washes performed in ice‐cold solutions, and fetal erythroblasts were mixed with adult erythrocytes (1:10 000) to make weak differential positive staining obvious. Secondary antibody was omitted in negative controls, while white blood cells acted as positive controls.

Surface charge

A slide‐size, free‐flow microelectrophoretic chamber constructed of acrylic plastic was mounted on an inverted microscope. Stainless steel electrodes generated an electric field of 3.6 V/mm, of reversible polarity. Electrolysis and thermal currents were minimized by use of an osmotically balanced sucrose buffer (0.25 mol/l) prepared in pure water (18 MΩ/cm at 25°C) (Milli‐Q systems, Millipore, USA).

Cells suspended in 1 µl phosphate‐buffered saline (PBS) were pipetted into the chamber, and migration events observed in real time were documented using manual time‐lapse photography. Cell migration was confirmed to be electrically generated by alternating the polarity of the electrodes. In test samples of adult erythrocytes, this induced instantaneous reversal of motion towards the new anode. Polarity of charges on first trimester fetal (primitive) erythroblasts, adult white and red blood cells were first determined individually, and subsequently confirmed in mixtures of fetal NRBC in maternal blood (1:1000 nucleated cells).

Buoyant density

A total of 2×10⁶ nucleated cells from four fetal blood samples at 9⁺² to 11⁺⁰ weeks were diluted in PBS to 200 µl; control samples of 2×10⁶ maternal erythrocytes were similarly prepared. Five commonly used Ficoll density gradients (Sigma) (1077, 1083, 1107, 1110, 1119) and 14 Percoll gradients between 1119 and 1140 were prepared (Amersham Pharmacia Biotech, UK). Cells (2×10⁶) diluted in PBS were layered over 1 ml Percoll or Ficoll in 1.5 ml polypropylene tubes and centrifuged at 654 g for 30 min. Cells were counted both in the gradient interface and the pellet.

In subsequent experiments, 10⁴ fetal erythroblasts were mixed into each 3 ml sample of male peripheral venous blood, diluted 1:1 in PBS. This was layered over an equal volume of Percoll (1117, 1118 or 1119) or Ficoll 1119 in 15 ml polypropylene tubes and centrifuged at 454 g for 30 min. Proportions of fetal erythroblasts recovered from four experiments for each of the four density gradients were analysed.

Susceptibility to ammonium chloride lysis

Fetal primitive erythroblasts were mixed with adult erythrocytes (1:10 000) in 50 µl PBS and exposed to four dilutions of ice‐cold ammonium chloride for 30 min (4:1, 10:1, 15:1 and 20:1 v/v). Carbonic anhydrase activity was blocked with 1 mmol/l acetazolamide and lysis of both cell types quantified at 20 and 30 min. Percentages of lysed cells were calculated over an average of three repeat experiments. Recovered fetal primitive erythroblasts were immunostained for ϵ‐globin (Europa Bioproducts, UK) and underwent XY cFISH (Choolani et al., 2001), as reported previously.

Immunomagnetic cell sorting with anti‐GPA, anti‐CD45 and anti‐CD71

Fetal nucleated erythrocytes (10⁵) between 8⁺⁰ and 10⁺⁶ weeks were labelled with either anti‐GPA, anti‐CD45 or anti‐CD71 antibodies conjugated to Miltenyi‐MACS beads (Miltenyi Biotech, Germany, GMBH). Magnetic separation was performed using miniMACS columns. Unlabelled cells sorted in the same way were used as controls for cell loss due to inherent process inefficiency.

Mixtures of first trimester fetal erythroblasts in maternal blood

Male fetal erythroblasts (9⁺⁰ to 11⁺⁰ weeks) were mixed with white blood cells retrieved from Ficoll 1119 of female non‐pregnant adult blood (25 000:1 000 000 cells; 1:40). Four sets of mixtures underwent magnetic separation for fetal NRBC, each set comprising four fractions: (i) CD45–: negative selection (depletion) using anti‐CD45; (ii) CD71+: positive selection using anti‐CD71; (iii) GPA+: positive selection using anti‐GPA; and (iv) CD45–/GPA+: CD45 depletion followed by positive selection using anti‐GPA.

Sorting first trimester fetal erythroblasts from maternal blood

Nine maternal blood samples from apparently normal pregnancies were studied: six obtained <5 min post‐termination and three before termination of pregnancy. Trophoblast analysis confirmed that all but one fetus was male. The case with the female fetus had maternal blood obtained after pregnancy termination. A tenth sample of maternal blood was obtained a week after chorion villus biopsy, but before the termination of a pregnancy complicated by trisomy 13. Three control blood samples from non‐pregnant adults were sorted and examined by Wright’s staining. The protocol designed for sorting these peripheral blood samples was based upon the characteristics of fetal primitive erythroblasts that had been studied; of several possible combinations, one was chosen.

Maternal blood (35 ml) from apparently normal pregnancies underwent density gradient centrifugation with Percoll 1118. White blood cells were depleted using anti‐CD45 and anti‐CD14 (Busch et al., 1994), and red blood cells were selected using anti‐GPA. The red cells were smeared or centrifuged onto positive‐charged glass slides and Wright‐stained for morphology and labelled for ϵ‐globin with AMCA (7‐amino‐4‐methylcoumarin‐3‐acetic acid) and underwent XY cFISH (Choolani et al., 2001). Maternal blood obtained from the patient carrying the trisomy 13 fetus was processed and analysed in the same way, except that enriched cells were analysed by cFISH for trisomy 13 (AneuVysion; Vysis, USA).

Statistics

Parametric data are presented as mean and 95% confidence intervals (CI = ±1.96 SE), and the differences between groups were determined using the two‐tailed Student’s t‐test or analysis of variance with Tukey post‐hoc analysis. Non‐parametric data are presented as median and range, and the differences between groups determined using the Mann–Whitney, χ², Fisher exact or Kruskal–Wallis tests. Statistical analyses were performed using SPSS Software (SPSS Inc., USA).

Results

Concentration of primitive and definitive erythroblasts

The mean total nucleated cell concentration of fetal blood remained similar between 8 and 12 weeks, after which it fell sharply: 8⁺⁰–9⁺⁶ weeks: 80.5×10⁶/ml (95% CI, 67.3–93.7; n = 17); 10⁺⁰–11⁺⁶ weeks: 93.3×10⁶/ml (95% CI, 67.8–118.8; n = 12); and 12⁺⁰–13⁺⁶ weeks: 20.4×10⁶/ml (95% CI, 14.9–26.0; n = 7). Proportions of erythroblasts in the same gestational ranges were 96.4% (95% CI, 95.1–97.7%), 93.8% (95% CI, 92.7–94.9%) and 90.6% (95% CI, 89.3–91.9%) respectively. Total erythroblast concentration in fetal blood was significantly lower after 12 weeks than before 12 weeks (z = 4.0; n = 36; P < 0.001). The frequency of morphological primitive erythroblasts fell progressively across the first trimester to reach negligible levels by 14 weeks (y = 252 – 16.7x; R² = 0.7; n = 14; P < 0.001). There was a reciprocal rise in the frequency of definitive NRBC across the same gestational age range. This regression equation is very similar to that previously reported for ϵ‐globin‐positive erythroblasts with gestational age (Choolani et al., 2001). This is because 100% of primitive erythroblasts expressed ϵ‐globin. In contrast, only one definitive NRBC was ϵ‐globin‐positive amongst 1800 (<0.06%) examined on nine representative slides between 9 and 12 weeks gestation (Fisher exact test: P = 0.001).

ϵ‐globin positive anucleate erythrocytes were observed only rarely within pure first trimester fetal blood samples (Figure 1A). Since we planned to use ϵ‐globin as the fetal cell identifier for first trimester fetal NRBC enriched from maternal blood, subsequent experiments focused mainly on primitive erythroblasts.

Surface antigen expression

All first trimester fetal erythroblasts stained positive for GPA and CD47 and negative for CD45 and CD35; all primitive erythroblasts were CD36– whereas all definitive erythroblasts were CD36+ (Figure 2B). CD71 expression on first trimester erythroblasts (mean, 75.3%; 95% CI, 66.5–84.1; n = 14) was lower than that reported for second trimester erythroid cells (Zheng et al., 1997), due to low or undetectable CD71 expression by primitive erythroblasts (Figure 2C). Only 68.0% (median; range: 0–87.5%) of primitive erythroblasts were CD71+ compared with 100% of definitive erythroblasts (z = 3.4; P = 0.001). Not only were there fewer CD71+ primitive erythroblasts, the expression on positive cells was also weaker than on definitive NRBC [mean pixel intensity 140.8 AU; 95% CI, 114.1–167.5; n = 5 and 78.8 AU; 95% CI, 71.1–86.4; n = 5; F(3, 16) = 76.8; P < 0.001 respectively].

Surface charge

Fetal primitive erythroblasts migrated towards the anode. Reversal of electrode polarity caused immediate reversal in the direction of travel towards the new anode, confirming their negative charge. Although primitive NRBC migrated more slowly than adult white and red blood cells, in mixing experiments, target primitive erythroblasts could not be separated effectively on this basis from maternal blood cells.

Buoyant density

Fetal erythroblasts in first trimester whole blood had buoyant densities ranging from 1.077 to 1.130 g/ml. As density increased, more fetal erythroblasts were recovered from the gradient interface and fewer cells settled in the pellet. In single density gradient experiments using Ficoll 1077, 1083, 1107, 1110 and 1119, 91.4%, 82.3%, 76.5%, 70.8% and 68.3% respectively (y = 616 – 0.5x; R² = 0.9; P < 0.001) of NRBC settled in the pellet. In single gradient experiments using Percoll 1119–1140, graduated at 0.001 g/ml intervals, no erythroblast (primitive or definitive) settled in the pellet when the gradient density was >1.130 g/ml (Figure 3). However, in less dense gradients, an increasing percentage of cells settled in the pellet with an inverse fall in the percentage retrieved from the gradient interface (y = 5.5x – 6138; R² = 0.7; n = 12; P = 0.002); overall cell loss was constant across all gradients (mean 18.9%; 95% CI, 11.6–26.2; R² = 0.1; n = 12; P = 0.26). In control samples (2×10⁶ maternal erythrocytes), although there was no clear separation between the densities of fetal erythroblasts and maternal anucleate red blood cells, ≥96.7% of maternal erythrocytes settled in the pellet if the density of the gradient was <1.124 g/ml. Thus, these data from single density gradient experiments of pure samples of fetal erythroblasts and maternal erythrocytes suggested that using a gradient density <1.124 g/ml, ≥96.7% of maternal anucleate red blood cells would settle in the pellet whereas up to 45.0% of fetal (primitive and definitive) NRBC could be recovered from the gradient interface (Figure 3).

To select the optimal density gradient <1.124 g/ml, the behaviour of adult peripheral blood diluted 1:1 in PBS was determined when centrifuged over Percoll gradients between 1.117 and 1.125 g/ml. Only Percoll 1117–1119 gave distinct layers at gradient interfaces, whereas Percoll 1120–1125 resulted in smears of cells unsuitable for collection (data not shown). Thus, Percoll 1117, 1118 and 1119 were used in subsequent experiments where fetal (mostly primitive) erythroblasts (9⁺⁰ and 11⁺⁰ weeks) were mixed with maternal blood and the yield of fetal cells determined. Recovery of fetal primitive erythroblasts using these three Percoll densities was compared with the yield from Ficoll 1119 (Troeger et al., 1999; Samura et al., 2000b; Prieto et al., 2001).

Median recovery of fetal primitive erythroblasts from mixtures in maternal blood was 64.1% (range: 58.1–71.8; n = 4) using Percoll 1118 compared with 10.1% (range: 8.3–13.6; n = 4) using Percoll 1117, 5.5% (range: 2.9–6.1; n = 4) using Percoll 1119 and 35.3% (range: 28.3–41.0; n = 4) with Ficoll 1119 (Table I). Percoll 1118 was superior to Percoll 1117 or 1119 (χ² = 8.0; df = 2; P = 0.02) and Ficoll 1119 (z = 2.3; P = 0.02). Fetal primitive erythroblasts were also found within mature anucleate red cell pellets. This loss of fetal primitive erythroblasts is inevitable since many fetal primitive erythroblasts have the same density as adult anucleate erythrocytes. However, Percoll 1118 limits loss into the red cell pellet and allows recovery of up to 71.8% of fetal erythroblasts.

Effect of ammonium chloride

Selective lysis of adult red cells in mixtures of fetal primitive erythroblasts showed that the latter lysed only after 15:1 or 20:1 (v/v) exposure to ammonium chloride/1 mmol/l acetazolamide for 30 min. The percentage of fetal NRBC lysed was 3.2 and 5.9% respectively. Maximal lysis of adult erythrocytes (87.5%) without lysis of fetal primitive NRBC was obtained by exposing cell mixtures to 10:1 v/v ammonium chloride/1 mmol/l acetazolamide for 30 min. Significant clumping between unlysed erythrocytes and primitive NRBC made it impossible to recognize target cell morphology. Thus, while selective lysis of adult erythrocytes could be useful in conjunction with diagnostic techniques such as PCR, where morphological identification of the target cells is less important, it is not suitable for non‐invasive prenatal diagnosis where ϵ‐globin‐positive primitive erythroblasts need to be identified by their characteristic morphology.

Anti‐GPA, anti‐CD45 and anti‐CD71 cell sorting

When pure fetal blood was sorted with anti‐GPA, anti‐CD45 or anti‐CD71, mean recovery of primitive erythroblasts was greatest using anti‐GPA, intermediate with anti‐CD45 and poorest with anti‐CD71 [F(2, 9) = 222.7; P < 0.001] (Table II). The mean recovery using positive selection with anti‐GPA (95.4%) was significantly greater compared with negative selection (depletion) using anti‐CD45 (88.1%; Tukey 95% CI of difference, 1.8–12.7; P = 0.01). Positive pressure elution of cells trapped in the MACS columns probably accounted for this advantage. Cell loss due to process inefficiency in all three groups was similar to that observed when unlabelled cells were passed through the columns (Table II).

Mixtures of fetal NRBC in maternal mononuclear cells (1:40) were separated into four fractions: (i) CD45–; (ii) CD71+; (iii) GPA+; and (iv) CD45–/GPA+. When sorting primitive erythroblasts from mixtures in maternal blood, fetal cell recovery was poorest using positive selection with anti‐CD71 (P < 0.001; Table III). Positive selection with anti‐CD71 gave poor recovery (38.7%) and low purity (38.0%), and 30.0% of target cells were retained in the discarded fraction. Anti‐CD45 depletion achieved high yields (75.9%) but purity was only 87.0%, with 10.0% of target cells retained in the column with the discarded cells. Fetal cell recovery with positive selection using anti‐GPA was lower than CD45 depletion (67.0%) but 100% purity was achieved and only 2% of target cells were lost. Anti‐CD45 depletion followed by anti‐GPA enrichment successfully combined the advantages of both strategies, with high yields similar to CD45– alone (76.0%) and the 100% purity of GPA+. Thus, this strategy was selected to enrich primitive erythroblasts from clinical maternal samples.

Sorting first trimester fetal erythroblasts from maternal blood

Nine maternal blood samples between 8⁺⁰ and 10⁺⁶ weeks were studied. Fetal primitive erythroblasts were identified by Wright’s staining in the first three maternal blood samples taken immediately post‐termination (Figure 2D). The numbers of target cells retrieved were 112, 1098 and 19 per 35 ml maternal blood. In contrast, no primitive erythroblasts were enriched from peripheral venous blood of non‐pregnant adults. For the second set of three post‐termination maternal samples, fetal gender was confirmed only after cell sorting and staining for ϵ‐globin and cFISH so that gender was predicted prospectively in a blinded fashion. The numbers of target cells (fetal primitive erythroblasts) retrieved were 27, 9 and 14 per 35 ml maternal blood. All primitive fetal erythroblasts were ϵ‐globin‐positive and all ϵ‐globin‐positive cells were primitive fetal erythroblasts. Predicted fetal gender matched cFISH on trophoblast: two male and one female fetus (Figure 2E, F). In the three pre‐termination maternal blood samples from apparently normal pregnancies, only two ϵ‐globin‐positive XY fetal primitive erythroblasts were identified, both from the same maternal blood sample (Figure 2G). No ϵ‐globin‐negative XY cells or ϵ‐globin‐positive XX cells were identified in any pre‐termination maternal sample. In contrast, four ϵ‐globin‐positive trisomy 13 fetal primitive erythroblasts were identified from the pre‐termination sample of maternal blood obtained from the patient carrying the trisomy 13 fetus (Figure 1H).

Discussion

This study describes characteristics of first trimester fetal primitive erythroblasts that may be applied to the development of fetal cell enrichment protocols for early non‐invasive prenatal diagnosis. The antigenic profile of first trimester fetal primitive erythroblasts was similar to adult erythrocytes, being GPA+ and CD47+ (100%) and CD35–, CD36– and CD45– (100%). However, in contrast to adult NRBC, 96% of which strongly express CD71 (Loken et al., 1987), only 68% of primitive erythroblasts were positive, and then only weakly. Although the morphologically heterogeneous populations of nucleated cells within fetal blood had densities ranging from 1.107 to 1.130 g/ml, the best fetal cell enrichment from mixtures in maternal blood was obtained using Percoll 1118. Primitive erythroblasts carried a net negative surface charge and were resistant to ammonium chloride lysis, but neither characteristic alone could effectively separate them from other cells in maternal blood. In accord with their surface antigens, MACS sorting of primitive erythroblasts was most efficient using anti‐CD45 and anti‐GPA, and least efficient using anti‐CD71, the most frequently used antibody for enriching fetal erythroblasts from maternal blood (Price et al., 1991; Cheung et al., 1996; Durrant et al., 1996; Troeger et al., 1999; Zhong et al., 2000; Al‐Mufti et al., 2001). Based on the above characteristics (buoyant density, surface antigen profile and resistance to ammonium chloride lysis) we developed a protocol to demonstrate enrichment of morphologically distinctive, ϵ‐globin‐positive fetal primitive erythroblasts from maternal blood between 8 and 13 weeks of pregnancy.

Fetal primitive erythroblasts share antigenic properties with both NRBC and mature anucleate erythrocytes. Poor expression of CD71, although surprising, supports the hypothesis that circulating primitive NRBC represent a terminally differentiated population of erythroid cells (Kelemen et al., 1979). Presence of ϵ‐positive primitive erythroblasts alongside uniformly ϵ‐negative anucleate erythrocytes is indirect evidence that these cells do not extrude their nuclei, but instead are cleared from the circulation as nucleated erythrocytes. Poor expression of CD71 accounts for their limited recovery in sorting experiments to date and explains why protocols using anti‐CD71 found the optimal gestational age for enriching fetal erythroblasts to be 15 weeks (Rodriguez De Alba et al., 2001). Although fetal erythroblasts have been enriched from second trimester maternal blood using anti‐CD36 (Bianchi et al., 1993, 1994; Campagnoli et al., 1997; Sohda et al., 1997; Troeger et al., 1999), we found that first trimester primitive erythroblasts were CD36–. Similar antigen expression precludes CD35 being used to separate fetal erythroblasts from adult erythrocytes. Uniform expression on both cell types similarly precludes using CD47, and also suggests that CD47 recognition is unlikely to be responsible for fetal cell clearance from maternal blood (Sekizawa et al., 2000; Kolialexi et al., 2001). Thus, in the absence of any fetally specific surface antigen, enrichment strategies need to be based on antigens that are present, or absent, in 100% of target cells. These are GPA and CD45. White cell depletion, either alone or prior to positive selection with GPA, improved yield (P = 0.02) while positive selection with GPA enhanced purity (P = 0.002).

The net surface negative charge on fetal primitive erythroblasts was inadequate to allow separation of these cells from adult red and white blood cells. Our data do not substantiate reports of using charge alone to separate fetal NRBC from maternal blood, at least in the first trimester (Wachtel et al., 1996, 1998; Shulman et al., 1998). However, our data support the use of positively charged slides for collecting the products of enrichment protocols (Wang et al., 2000).

Protocols developed using cord blood NRBC in model mixtures to optimize density gradient enrichment usually recover fetal erythroblasts from maternal blood in the second trimester (Prieto et al., 2001; Rodriguez De Alba et al., 2001). Recent experiments have suggested the use of denser gradients and that Ficoll 1119 enhances fetal cell recovery (Troeger et al., 1999; Samura et al., 2000b). Given the heterogeneous population of nucleated cells within fetal blood, it is not surprising that fetal erythroblasts have been enriched from maternal blood by a variety of density gradients. Primitive erythroblasts have not been recovered from first trimester maternal blood because the majority (68.3%) settle in the pellet when Ficoll 1119 is used. We made the surprising observation that Percoll 1118 gave 1.8‐fold greater recovery of fetal primitive erythroblasts compared with Ficoll 1119. The reason for this difference is unclear although the manufacturers indicate that the two solutions may behave differently. Percoll 1118 allowed retention of primitive erythroblasts within the nucleated cell fraction at the gradient interface, which can then be isolated by immunomagnetic cell sorting.

Fetal primitive erythroblasts were resistant to ammonium chloride lysis but unexplained changes in the membrane adhesion properties of the remaining unlysed fetal and adult red cells caused significant clumping and poor cell morphology (Webster and Pockley, 1993; Ridings et al., 1996). Although selective erythrocyte lysis has been used in fetal cell recovery (Boyer et al., 1976; De Graaf et al., 1999), recent data (Voullaire et al., 2001) support our suggestion that ammonium chloride lysis has only a limited role in fetal erythroblast enrichment.

Proof‐in‐principle that a protocol based on the characteristics of first trimester primitive erythroblasts could successfully enrich these target cells from maternal blood was initially obtained in the frequently used biological model of feto‐maternal haemorrhage in post‐termination maternal blood (Sekizawa et al., 1999; Samura et al., 2000a; Wang et al., 2000; Bianchi et al., 2001). After 1118 Percoll, CD45 and CD14 depletion, and after GPA enrichment, fetal primitive erythroblasts were identified by their characteristic morphology and ϵ‐globin positivity. Their fetal origin was confirmed by gender concordance with fetal tissue. Enrichment of primitive erythroblasts was less reliable in maternal blood samples obtained before termination of apparently normal pregnancies, but finding two ϵ‐globin‐positive XY primitive erythroblasts in one of the three pre‐termination samples tested confirms that these cells circulate in first trimester maternal blood. Identification of four trisomy 13, ϵ‐globin‐positive, primitive erythroblasts from the single pre‐termination maternal blood sample that was obtained from the patient carrying the trisomy 13 fetus demonstrates its potential clinical application to non‐invasive prenatal diagnosis in the first trimester. These findings are in keeping with observations by others (Bianchi et al., 1997) that suggest very few fetal cells circulate in maternal blood, but the number increases in aneuploid cases. Our low success rate in euploid pre‐termination samples may reflect a biological limit to physiological feto‐maternal haemorrhage in early pregnancy, or alternatively loss of fetal cells during the sorting process.

It is now accepted that the first trimester of pregnancy (Cheung et al., 1996) is the clinically‐preferable time for non‐invasive prenatal diagnosis. To date, however, limited access to first trimester fetal blood has hindered development of enrichment techniques based upon specific biological properties of these target cells. Recently, investigators have studied fetal NRBC in chorionic villus sample washings but such samples are often contaminated by maternal blood cells and tissue (Mesker et al., 1998; Mavrou et al., 1999; Jakobs et al., 2000; Voullaire et al., 2001). Our study confirms ϵ‐globin as a very specific marker for fetal primitive erythroblasts (Boye et al., 2001; Choolani et al., 2001; Hogh et al., 2001; Voullaire et al., 2001) by showing that this embryonic globin is present in 100% of primitive erythroblasts, but only rarely (<0.06%) in definitive NRBC. To improve enrichment from first trimester maternal blood, we studied the biological properties of primitive erythroblasts so as to develop a protocol for their selection from, and identification in, first trimester maternal blood. Development and testing of different enrichment regimens, based on these properties in intact pregnancies, is now a priority for first trimester non‐invasive prenatal diagnosis using fetal primitive erythroblasts derived from maternal blood.

Acknowledgements

We would like to thank M.Antoniou for providing the K562 cell line. M.C. is supported by the Overseas Graduate Scholarship from the National University of Singapore and K.O.D. by a grant from Action Research. Consumables were additionally supported by grants from the Hammersmith Hospital Trust Research Committee and the Institute of Obstetrics and Gynaecology Trust (Registered Charity No. 292518).

References