Centrifugation affects the purity of liquid biopsy‐based tumor biomarkers

Abstract

Biomarkers in the blood of cancer patients include circulating tumor cells (CTCs), tumor‐educated platelets (TEPs), tumor‐derived extracellular vesicles (tdEVs), EV‐associated miRNA (EV‐miRNA), and circulating cell‐free DNA (ccfDNA). Because the size and density of biomarkers differ, blood is centrifuged to isolate or concentrate the biomarker of interest. Here, we applied a model to estimate the effect of centrifugation on the purity of a biomarker according to published protocols. The model is based on the Stokes equation and was validated using polystyrene beads in buffer and plasma. Next, the model was applied to predict the biomarker behavior during centrifugation. The result was expressed as the recovery of CTCs, TEPs, tdEVs in three size ranges (1–8, 0.2–1, and 0.05–0.2 μm), EV‐miRNA, and ccfDNA. Bead recovery was predicted with errors <18%. Most notable cofounders are the 22% contamination of 1–8 μm tdEVs for TEPs and the 8–82% contamination of <1 μm tdEVs for ccfDNA. A Stokes model can predict biomarker behavior in blood. None of the evaluated protocols produces a pure biomarker. Thus, care should be taken in the interpretation of obtained results, as, for example, results from TEPs may originate from co‐isolated large tdEVs and ccfDNA may originate from DNA enclosed in <1 μm tdEVs. © 2018 The Authors. Cytometry Part A published by Wiley Periodicals, Inc. on behalf of International Society for Advancement of Cytometry.

Potential biomarkers in the blood of cancer patients include circulating tumor cells (CTCs) 1, 2, 3, tumor‐educated platelets (TEPs) 4, tumor‐derived extracellular vesicles (tdEVs) 5, EV‐miRNAs 6, and circulating cell‐free DNA (ccfDNA) 7, 8. Typically, the first step to isolate a biomarker involves (differential) centrifugation, which isolates particles based on size and density. After centrifugation and optionally further processing, the biomarker is measured based on antigen exposure (CTCs and tdEVs) or on the composition of RNA (TEPs and EV‐miRNAs) or DNA (ccfDNA). Fractions containing the biomarker of interest may be impure and contain substantial quantities of “contaminants”, for example, other biomarkers. These contaminants may affect the obtained signal and potentially lead to the misinterpretation of results. Therefore, we applied a model based on the Stokes equation to assess the purity of biomarkers after centrifugation according to protocols used by other groups to study their biomarkers of interest 1, 2, 3, 4, 5, 6, 7, 8.

Materials and Methods

Stokes Model

Figure 1 shows the principle of centrifugation. Before centrifugation, particles are uniformly distributed in a medium. Upon centrifugation, particles denser than the medium will travel toward the bottom of the tube. After centrifugation, the top fraction is collected, defined here as the “supernatant.” We define the “pellet” as the fraction remaining in the tube. The dashed line in Figure 1 is the interface between the supernatant and the pellet. The recovery of particles in the pellet is the fraction of particles in the pellet after centrifugation. This recovery depends on the starting point of the particle in the medium, as well as the densities and diameters of the particles. Consequently, the pellet will contain a high concentration of large and high density particles compared to the supernatant. For a swing‐out rotor, the distance traveled by a particle can be modeled by the Stokes equation. In the Stokes equation, the speed of the particle results from the balance between the buoyant force and the drag force on a particle in a medium. In a centrifuge, the gravitational acceleration g depends on the distance from the axis of rotation.

$$g=R{\omega }^2$$ (1)

Figure 1.

Centrifugation to isolate biomarker of interest from whole blood. The position of a particle after centrifugation can be described with the stokes equation (Eq. (2)). Centrifugation of particles that are uniformly distributed throughout the sample causes large, high‐density particles to move down to the pellet (below the dashed line), while small, low‐density particles stay in the supernatant (above the dashed line).

where R is the distance to the axis of rotation and ω is the angular velocity of the centrifuge. After time T, the distance R of a particle starting at R ₀ is given by the following equation:

$$R_{\mathrm{T}}=R_0{e}^{\frac{d^2\left({\rho }_{\mathrm{P}}-{\rho }_{\mathrm{M}}\right)}{18S\eta }{\omega }^{2}T}$$ (2)

where d and ρ _P are the diameter and the volumetric mass density of the particle and ρ _M and η are the volumetric mass density and viscosity of the medium. To allow nonspherical shapes such as platelets, the shape factor S is introduced (S = 1 for spheres). S accounts for the additional drag these nonspherical particles experience 9. Note that for nonspheres, d is the diameter of a sphere with equivalent volume. Model assumptions are (1) negligible wall effects, (2) instantaneous formation of the packed cell fraction during centrifugation of whole blood, and (3) aspiration sufficiently gentle not to disturb the pellet (see 10, 11 for a complete description of the assumptions in the Stokes model). The model was programmed in Matlab (v2017a; MathWorks, Natick, MA) and applied to the evaluated centrifugation protocols.

Stokes Model Sample Properties

Table 1 shows the assumed sample properties, which were derived from the literature for a temperature of 20°C (see Supporting Information Table S1 for additional background). If available, volumetric mass densities were taken from references that applied density media with neutral osmotic effects. Platelets have a log‐normal volume distribution 12 and a shape factor S of 2.0 9, 13. Platelets can have a volumetric mass density of 1.05–1.09 g/ml, where the lower density is associated with platelets that have secreted their α‐granule content but with an unaffected platelet volume distribution 14, 15. Most EVs have a diameter of <200 nm, which means that the volumetric mass density is substantially affected by the membrane density. A model describing EVs as cytoplasm enclosed by a cell‐like membrane 16 leads to a size‐dependent volumetric mass density. For EVs 100, 200, and 1,000 nm, the density is 1.099, 1.081, and 1.064 g/ml, respectively, which is in fair agreement with literature estimates between 1.08 and 1.11 g/ml 17, 18, 19, 20, 21. Finally, the mass density of ccfDNA is only available in cesium chloride media and the length of ccfDNA is unknown. Assuming a length of <1,000 base pairs, the ccfDNA would be smaller than 50 nm.

Table 1.

Stokes model sample properties

Type and property	Value used in model	References
CTCs
Density (g/ml)	1.053	24, 25
Diameter (μm)	Uniform. 8–20	26
Platelets
Density (g/ml)	Norm., mean 1.069, SD 0.0053	14, 15, 27
Diameter (μm)	Lognorm., median 2.4, SD 0.6	12, 28
Shape factor	2.0	9, 13
EVs
Density (g/ml)	Core 1.060, 8 nm shell 1.155	17, 18, 19, 20, 21
Diameter (μm)	Uniform. 0.05–1
ccfDNA
Density (g/ml)	1.7	16, 29
Diameter (μm)	≤0.05	16
Polystyrene beads
Density (g/ml)	1.05	MFG
Diameter (μm)		MFG
PBS
Density (g/ml)	1.004	30
Viscosity (mPa s)	1.193	30
Plasma
Density (g/ml)	1.0253	31, 32
Viscosity (mPa s)	1.75	33, 34, 35

Values at 20°C (see Supporting Information Table S1 for more details). Lognorm., lognormal distribution; MFG, manufacturer specifications; Norm., normal distribution; SD, standard deviation; Uniform., uniform distribution.

Model Validation

To validate the model, polystyrene beads were diluted in phosphate‐buffered saline (PBS) or blood plasma. The concentration of beads was measured by flow cytometry before and after centrifugation at 300g for 20 min or 2,700g for 22 min using a Rotina 380R centrifuge (Hettich, Tuttlingen, Germany) or 15,800g for 60 min using a SW 41 Ti rotor and Optima L‐80 XP ultracentrifuge (Beckman Coulter, Fullerton, CA), all at 20°C and with deceleration set to the minimum possible value. Polystyrene beads were 400, 799, 994, and 3,005 nm in diameter (Thermo Fisher Scientific, Waltham, MA) with a density of 1.05 g/ml. PBS (154 mM NaCl, 1.24 mM Na₂HPO₄·2H₂O, 0.2 mM NaH₂PO₄·2H₂O; pH 7.4) was 0.22 μm filtered (Merck, Darmstadt, Germany). Blood from anonymous healthy donors was obtained after written informed consent in accordance with the Dutch regulations and approved by the medical–ethical assessment committee of the Academic Medical Center, University of Amsterdam. Whole blood was drawn using a 21G needle in EDTA vacutainers (BD Biosciences, San Jose, CA) and processed within 15 min after collection. Plasma was obtained from whole blood by double centrifugation at 2,500g for 15 min at 20°C using a Rotina 380R centrifuge. The supernatant was pooled between centrifugation steps and pooled before mixing with the polystyrene beads. The concentration of beads before and after centrifugation was measured on side scatter of a flow cytometer (FACSCalibur; BD, Franklin Lakes, NJ, 15 mW 488 nm laser, flow rate ~60 μl/min calibrated by weight 22, SSC 400 V, gain 0, threshold 0), and data were analyzed using FlowJo (v10; FlowJo LLC, Ashland, OR).

Predicted Performance of Centrifugation Protocols

We modeled the protocols as described in the literature and shown in Table 2. The dimensions of the low‐speed centrifugation tube were derived from the 57.462 Sarstedt centrifugation tube and for ultracentrifugation from the 344059 Beckman Coulter centrifugation tube. For the model, we assume that centrifugation <10,000g is performed using a Rotina 380R centrifuge and centrifugation >10,000g using a SW41 Ti rotor and Optima L‐80 XP ultracentrifuge. For most centrifuges, the set g‐force is the force at the bottom of the tube. The force elsewhere is given by Eq. 1. This means that a 10 cm high sample in a Rotina centrifuge set to 2,500g, has a g‐force of 900g at the top of the sample. The same centrifugation protocol on a rotor with a larger diameter will increase the g‐force at the top of the sample and thus result in an increased recovery in the pellet. However, we did not model multiple rotor diameters, because most rotors suitable for large tubes have diameters comparable to the Rotina 380R. Results are expressed as the % recovery of CTCs, TEPs, tdEVs in three size ranges from large (1–8 μm, “apoptotic bodies”, “oncosomes”, or “tumor microparticles”), intermediate (0.2–1 μm, “microparticles”), to small (0.05–0.2 μm, “exosomes”) tdEVs, EV‐miRNAs, and ccfDNA.

Table 2.

Modeled centrifugation protocols

Protocol	Blood volume and vacutainer	Protocol summary	References
CTCs/tdEVs	10 ml CellSearch	WB 7.5 ml:6.5 ml PBS 800g, 10 min → (10) Pel	1, 2, 3, 5
TEPs	6 ml EDTA	WB 120g, 20 min → (3) Sup 360g, 20 min → (10) Pel, wash 2×	4
EV‐miRNA	6 ml EDTA	WB 900g, 7 min → (5) Sup 2,500g, 10 min → (5) Sup 500g, 10 min → (5) Sup	6
ccfDNA (Speicher)	10 ml PAXgene ccfDNA	WB 200g, 10 min, 1,600g, 10 min → (3) Sup 1,600g, 10 min → (3) Sup	8
ccfDNA (Dawson)	10 ml EDTA	WB 820g, 10 min → (5) Sup 20,000g, 10 min → (10) Sup	7

ccfDNA, circulating cell‐free DNA; CTCs, circulating tumor cells; EV‐miRNAs, extracellular vesicle‐associated miRNAs; TEPs, tumor‐educated platelets; tdEVs, tumor‐derived extracellular vesicles; WB, whole blood; → (xx) Pel, collect pellet xx mm above bottom of tube or buffy coat (for WB); → (xx) Sup, collect supernatant xx mm above bottom of tube or buffy coat (for WB).

Results

Model Validation

Figure 2 shows the modeled and measured bead recovery in PBS (top) and plasma (bottom). As expected, the recovery of beads in the pellet increases with increased particle diameter, g‐force, and centrifugation time. In addition, because plasma is more viscous than PBS, the recovery of beads in the pellet is lower in plasma compared to PBS. Modeled and measured recovery are in fair agreement (all errors <8%) for both PBS and plasma at 300g and 2,700g, as well as for PBS at 15,800g. The model overestimates the particle speed at 15,800g for plasma, with a maximum error of 20%. Thus, the model is a reasonable approximation to predict the behavior of spherical particles when applied to the centrifugation protocols as shown in Table 2.

Figure 2.

Centrifugation of beads diluted in PBS or plasma. The modeled bead recovery in pellet (blue dashed line) and supernatant (black, solid line) compared to measured bead recovery in pellet (open blue diamonds) and supernatant (closed black circles) for three centrifugation conditions.

Predicted Performance of Centrifugation Protocols

Table 3 shows the predicted recovery of number of particles based on the centrifugation protocols as shown in Table 2. The volume of the supernatant plus the volume of the pellet equals the starting volume. The volume reduction can be calculated by dividing the starting volume by the volume of supernatant or pellet. To obtain the particle concentration after centrifugation, the predicted recovery in the number of particles needs to be multiplied with the volume reduction, which is shown in the last column of Table 3.

Table 3.

Predicted recovery (%) of centrifugation protocols

Protocol	Change in number (% of original)						Vol. red.
	CTCs	TEPs	Large tdEVs	Intermediate tdEVs	Small tdEVs	ccfDNA	(fold)
CTCs/tdEVs	100.0	39.4	81.1	16.7	13.3	13.3	2.9
TEPs	0.0	71.1	21.7	3.2	2.2	2.2	3.0
EV‐miRNA	0.0	0.01	0.7	39.7	57.0	56.9	1.7
ccfDNA (Speicher)	0.0	0.05	2.4	63.4	82.0	82.0	1.2
ccfDNA (Dawson)	0.0	6.10−7	0.0	6.0	49.0	48.4	1.7

ccfDNA, circulating cell‐free DNA; CTCs, circulating tumor cells; EV‐miRNAs, extracellular vesicle‐associated miRNAs; TEPs, tumor‐educated platelets; tdEVs, tumor‐derived extracellular vesicles; Vol. red., volume reduction. Bold font highlights the target component.

The CTC protocol recovers 100% of CTCs but co‐isolates substantial fractions of TEPs, large tdEVs, and ccfDNA. In subsequent epithelial cell adhesion molecule (EpCAM)‐based magnetic enrichment of CTCs, the TEPs and ccfDNA will be removed, but tdEVs exposing EpCAM will be co‐isolated. The latter is confirmed by a previous study, in which the CTC fraction was shown to contain large EpCAM⁺ EVs 5.

The protocol to isolate TEPs predicts a yield of 71% together with 22% large tdEVs and <3% CTCs, smaller tdEVs, and ccfDNA. Because the subsequent processing includes detection of all mRNA present in the sample, our model indicates that additional evidence is needed to prove that the obtained RNA profiles indeed originate from platelets.

EV‐miRNA samples contain intermediate tdEVs (40%), small tdEVs (57%), ccfDNA (57%), and TEPs and large tdEVs (<1%). By subsequent size exclusion chromatography 23, ccfDNA is removed, resulting in a relatively pure tdEV sample with high yield. The model predicts that the concentration of tdEVs is unaffected by the final centrifugation step of the protocol, which is confirmed by the authors upon inquiry 6.

For ccfDNA, two centrifugation protocols were evaluated. The Speicher protocol 8 predicts the 82% recovery of ccfDNA, together with large tdEVs (2%), intermediate tdEVs (63%), small tdEVs (82%), and negligible CTCs and TEPs. The Dawson protocol 7 yields 48% of ccfDNA, but with lower contamination of intermediate tdEVs (6%), small tdEVs (49%), and negligible CTCs, TEPs, and large tdEVs. Thus, both protocols do not yield pure ccfDNA. Because the volumetric mass density difference between DNA and plasma is at least ninefold higher than the density difference for EVs, it is difficult to separate DNA from small EVs. Consequently, a 35 nm DNA “particle” travels at the same speed as a 100 nm EV.

Discussion

Centrifugation can be described with a model based on the Stokes equation. The model illustrates that centrifugation is effective at recovering particles that can be pelleted but that additional methods are needed to obtain a pure fraction. Particles with similar sedimentation rates are always co‐isolated during centrifugation. This is especially relevant for TEP and ccfDNA protocols, because the co‐isolated tdEVs may have substantial impact on the outcome. This model offers a tool to (re)design centrifugation protocols and at a minimum to establish whether available candidate protocols differ sufficiently from each other to warrant comparison. The model includes a shape factor to account for the nonspherical shape of platelets. This shape factor was set to 2.0 based on literature but was not experimentally validated.

For technologies evaluating the presence of CTCs, such as the CellSearch system, whole blood is centrifuged at 800g for 10 min. For this application, the model predicts a 100% effective isolation of CTCs, in addition to a recovery of 81% of the large tdEVs and 39% of TEPs. However, because of the use of imaging to identify each particle, tdEVs and TEPs can be distinguished from CTCs, and thus, the co‐isolation does not affect the determined CTC concentration. Nevertheless, a purer CTC sample may be obtained by adjusting the centrifugation protocol. To reduce the recovery of TEPs or intermediate size tdEVs, the g‐force or centrifugation time should be reduced. Because of their size, ccfDNA or small tdEVs’ recovery in the pellet is mainly reduced by a reduction of the collection height.

The model predicts that a single biomarker cannot be purified by centrifugation alone. Substantial fractions of co‐isolated other biomarkers remain present in all evaluated protocols, which may affect the obtained results and interpretation thereof.

Conflict of Interests

The authors declare that the research was carried out in the absence of any personal, professional, or financial relationship that could potentially be construed as a conflict of interest.

Supporting information

Table S1. Literature values for components in model.