Pipette cleaning in automated systems

Recently, a number of robotic sample handling systems have been introduced which perform sample preparation steps in laboratory tests. Many of these systems, such as those described by Severns and Hawk and Martin [2], have sophisticated positioning systems which are under computer control. Some of these robotic systems have been configured as dedicated pipettor/diluters whose positioning system manipulates a rigid tube, or pipette, which is connected to a computer-controlled syringe pump. Systems of this type are quite useful for introducing samples from sample tubes into secondary containers (such as microplates) for performing tests. In addition to minimizing the potential for pipetting errors, these systems can also perform data management functions, such as keeping track of sample identification and test results.


Introduction
Recently, a number of robotic sample handling systems have been introduced which perform sample preparation steps in laboratory tests. Many of these systems, such as those described by Severns and Hawk and Martin [2], have sophisticated positioning systems which are under computer control. Some of these robotic systems have been configured as dedicated pipettor/diluters whose positioning system manipulates a rigid tube, or pipette, which is connected to a computer-controlled syringe pump. Systems of this type are quite useful for introducing samples from sample tubes into secondary containers (such as microplates) for performing tests. In addition to minimizing the potential for pipetting errors, these systems can also perform data management functions, such as keeping track of sample identification and test results.
Many of the tests which are performed using automated pipetting systems are not sufficiently sensitive that carry-over between samples or tests introduces significant errors. Certain tests, however, such as the enzyme immunoassays (EIA and ELISA), radioimmunoassay (RIA) and fluorescent immunoassay (FIMA), are sensitive enough that cross-contamination may present a problem. One such assay is the screening procedure for Hepatitis B surface antigen (HBsAg) which is routinely performed on each unit of blood and plasma collected in the United States. Nath and Dodd [3] have reported that the sensitivity of this assay (which is available from various manufacturers as either an ELISA or an RIA) is better than ng of HBsAg per ml of sample, while the highest concentration of HBsAg found in samples is approximately mg per ml. Therefore, cross-contamination from sample to sample must be less than part per million iffalse positive results due to carry-over are to be avoided. Until recently, this stringent requirement for cross-contamination dictated the use of disposable pipette tips, adding substantially to the cost of assay. It was hoped that a method could be found to clean a non-disposable pipette adequately to perform this test.
One method which is frequently used to clean pipettes in automated systems is to place the tip of the pipette into a 'wash station' and to force liquid through it, as shown in figure 1. The liquid travels through the inside of the pipette, removing contaminants, then up the interior of the wash station, cleaning the outside of the pipette. A * Contribution No. 683 from the American Red Cross. substantial advantage of this approach is the low cost of water. The purpose of this study was to determine the primary factors which influence the removal of watersoluble contaminants from a stainless-steel pipette when distilled water is used as the cleaning liquid.

Methods and equipment
All studies were performed using a Tecan Sampler 505 (Tecan [US] Ltd, Hillsborough, North Carolina, USA). The sampler consists of a Cartesian (X,Y,Z) robotic arm and a pipette connected via Teflon tubing to a pair of motorized syringes. The pipette is a 130mm long stainless-steel tube of 2 mm outer diameter and 1.4 mm inner diameter. The last 20 mm of its length is narrowed, having an inner diameter of 0"4 mm. The outside of the pipette is coated with Teflon; the inside is not. A capacitance sensor allows the Sampler to detect the surface ofany conductive liquid. The entire assemblage is controlled by an IBM Personal Computer using a combination of RatBas [4]  An initial study, using coomassie blue as an analyte, was performed to determine whether contamination on the pipette was confined to its inside. It was determined that approximately 1"5 [1 of liquid adhered to the pipette, partitioned between the outside and the inside surfaces.
In the experiments which are described below, the cleanliness of the pipette is estimated by measuring the amount of analyte in liquid which is withdrawn from the wash station. As this liquid was just expelled from the pipette, but has also been in contact with the outside tip of the pipette, it should reflect the likelihood that contamination would be introduced into a subsequent sample.
To determine the effectiveness with which the pipette was being cleaned, the dilution (the ratio of measured concentration to the original concentration) of analyte in the liquid sampled from the wash station was analysed as a function of the volume of distilled water which had been dispensed to clean the pipette. The analytes (all of which were water soluble) were prepared in distilled water, 7% bovine serum albumin (BSA) or human serum. Because of the limited resolution of the measuring instruments, the analytes (except for HBsAg) were prepared in different initial concentrations to extend the range of dilutions which could be measured. The data from a single set of experimental conditions were pooled for analysis.
Several different analytes were used during this study. In all of the studies, except those utilizing HBsAg, each experiment was calibrated to determine the relationship 136 between the measured variable (absorbance or fluorescence) and the dilution; 24 serial two-fold dilutions were prepared by the Sampler. A technologist later added distilled water to each well so that the total volume was the same as in the actual wash procedures (250 1). The calibration thus consisted of dilutions from 2" to 16777216:1.
Each experiment was conducted on the same microplate as its calibration curve. For each wash volume, an air gap of 7 1 (to minimize mixing of the analyte with the wash solution) was first aspirated into the pipette. The Sampler then moved the pipette to the surface of the analyte (as determined by the liquid level sensor) and aspirated an aliquot which varied in volume depending upon the experiment. The Sampler was programmed to follow the surface of the liquid, remaining just sufficiently submerged (less than mm) to avoid aspirating air. The pipette was moved to the wash station, where the analyte and air gap were dispensed, followed by the proper volume of wash solution. A 50 1 aliquot ofliquid from the wash station was aspirated and placed into a well of the Because of changes in such items as protein concentration (when BSA or plasma was used as the carrier for the analyte) and pH (when MUB was used as the analyte) with dilution, there was often a nonlinear relationship between the measured variable (especially fluorescence) and dilution. To account for some of this variability, a second order polynomial was fitted to the calibration data using the method described by Forsythe [5]. The coefficients of the polynomial wer.e used to compute the dilutions from the experimental measurements.
After the experimental dilutions had been calculated, all data from a single set of experimental conditions were pooled. A minimum least squares fit of the appropriate model equation to the pooled data was determined using the Marquardt-Levenberg technique [6] or a pseudo Gauss-Newton technique described by Dixon [7]. As the expected form of the washout curves was a sum of exponentials, model equations were fit to the logarithm of the data points, so that all data points would influence the fit equally.
To determine which of two model equations best fit the data, the residuals from both models were examined to determine which explained a greater portion of the total variance. Residuals were first examined to see if they were normally distributed using the Wilk-Shapiro test. If both sets of residuals were normally distributed, their variances were tested using the F-test, otherwise they were tested using the Ansari-Bradley test. The application of these tests is described by Sokal and Rohlf [8].
To test data from two experimental conditions for equality, a two-way analysis of variance (ANOVA) [8] was performed on the data from the washout curves using experimental condition and wash volume as the classification variables. Parameter values derived from curve fitting were tested for equality using Student's t-test.
Standard error estimates from the model fit were used to approximate the standard deviation of the estimate, and the number of degrees of freedom was computed as N + N2 2, where N and N2 are the degrees of freedom from the parameter estimates.

Results
Effect of depth of submersion in the wash station  Additional experiments at various submersion depths indicate that the one compartment model is appropriate when the pipette is inserted at least 15 mm into the wash station cavity.
It requires significantly more liquid to attain a specified dilution when the tip of the pipette is submerged 5 mm than when it is submerged 25 mm (p < 0"0001, two-way ANOVA). In addition, the estimated compartmental volume is decreased from 234 1 (V 119 zl, V2 115 zl) to 87 zl (V1 87 zl). Thus, the pipette is cleaned in a more satisfactory fashion when its tip is placed near the bottom of the wash station. In the remainder of the experiments described, the tip of the pipette is placed of the wash stations, the differences in the estimated values of V were not statistically significant (p 0"17, two-way ANOVA). The estimate of the initial concentration was significantly lower in the smaller wash station (p < 0"001, two-way ANOVA). Thus, the depth of the wash station appears to have a small but definite effect on the washout of contaminants near the tip of the pipette.
Because both wash chambers had the same inside diameter, it was not possible to examine the effect of cross-sectional area.
Effect of volume of sample When a more sensitive analyte is used to study the washout process, it becomes apparent that the curve changes slope at high dilution. This change seems to reflect a second process which dominates the washout process for dilutions greater than approximately 1000: 1. Figure 3 shows typical experimental data generated using MUB as an analyte. The curve was fit using Equation 3 (Appendix A).

Effect of changing the wettability of the pipette surface
To determine if the wettability of the surface of the stainless-steel pipette affected the removal of contaminants, a stainless-steel pipette was exposed for 12 h to a 1% solution ofProsil-28 (PCR Research Chemicals, Inc., Gainesville, Florida, USA), an. organosilane surface treating compound, then air dried at 75 C for h. This treatment was found to substantially reduce the contact angle [9] of water droplets on stainless-steel, indicating that the surface had become considerably more hydrophobic. No difference was found in the washout of MUB between treated and untreated pipette (p > 0'5, two-way ANOVA). 138 To further verify the lack of effect of surface wettability, Tween-20 (polyethyleneoxidesorbitan laurate, Sigma Chemical Co., St. Louis, Missouri, USA), a surface wetting agent, was added to the distilled water which was flushed through the pipette. No change in the washout of MUB was detected (p > 0"7, two-way ANOVA).

Effect of the viscosity of the solution containing analyte
To determine ifthe viscosity ofthe solution containing the analyte affected the washout process (as it might if analyte is trapped in a film at the wall), fluorescein was used as an analyte in both distilled water (viscosity centipoise) and in 7% BSA (viscosity 1"8 centipoise). No difference was found in the washout curves for these two experimental conditions (p > 0"5, two-way ANOVA). To determine the effect of diffusion on the washout process, analytes of different molecular weights were used. Studies were performed using MUB (molecular weight 198"2) and HBsAg (approximate molecular weight 2'7 x 106). Table shows the estimates of V for these conditions; the estimates are significantly different (p < 0"001, t-test on parameters). It appears that the molecular weight of the analyte has a small but detectable effect on the washout process. Si.ce the coefficient of diffusion of a substance is (to a first approximation) 1072'0 related to the inverse of the square root of the molecular weight, the ratio of the diffusion coefficients of the two analytes is on the order of 100" 1. This change in diffusion coefficient resulted in a 2:1 change in the estimated value for V2.

Conclusions
The washout of contaminants from a pipette in a wash station (assuming that the tip of the pipette is close to the bottom of the wash station) can be modeled by two parallel well-stirred compartments. The response of the system consists of an initial (fast) portion and a later (slow) portion.
The initial part of the washout process appears to be related to the convective transport of analyte in the wash station itself. It is relatively unaffected by the amount of sample aspirated or by the diffusivity of the analyte; it is slightly affected by the depth of the wash station. The initial portion of the washout process appears to be most affected by the depth to which the pipette is inserted into the station. When the pipette is inserted only a small distance into the wash column, the flow in the column appears to be partitioned into at least two separate 'compartments', which do not mix well. It is possible that the partitioning reflects the existence of a plume ofliquid. When the pipette is lowered sufficiently into the wash column, the increased turbulence and decreased volume appear to cause better mixing.
The latter part of the washout process appears to be primarily related to the removal ofanalyte from the inside of the pipette. This portion of the washout process is significantly affected by the amount of sample which is aspirated and, to a lesser extent, by the flow rate and the diffusion coefficient of the analyte. When the flow rate was changed, the slope ofthe curve changed in a direction opposite to that predicted if diffusion were a dominant force in the washout process. It therefore appears that the washout of analyte during the second phase of the curve is primarily due to convective transport, and only secondarily due to diffusion of analyte. It is possible that the apparent effect of the molecular weight of the analyte is due to a non-diffusional binding effect at the wall. Further experiments with analytes of varying molecular weights may determine the nature of this effect.
Attempts were made to explain the washout of analyte using existing models of the dispersion of solute in a tube (Equations 4, 5 and 6, Appendix A). None of these models improve the explanation of the data, and the models for laminar flow with no diffusion (Equation 4), and turbulent flow (Equation 6) are significantly worse (p < 0"001, F-test).
The dependence of estimates of V2 on the volume of sample aspirated is striking. Figure 5 shows the relationship between the volume of sample aspirated and the estimate for the parameter V2 (Equation 3, Appendix A).
Since the measured interior volume of the pipette was 200 tl, the break which is visible in the curve may be due to the transition from the stainless-steel pipette to the Teflon connecting tubing. It appears that the change in slope may therefore be related to a change in the exposed surface area.
To determine if the predictions of the model system could be validated in practice, a pseudo-random sequence of 30 HBsAg positive samples and 30 negative controls were pipetted into an Abbott tray for the Auszyme II assay by the Tecan Sampler. A volume of 7 ml (predicted from the experiments above to adequately reduce carry-over) was used to wash the pipette between samples. No trace of carry-over was found in any of the negative samples. Appendix A Models of the washout process A number of standard models have been used to analyse data from the pipette cleaning experiments. The assumptions involved in formulating each of the models, and the equations which result, are briefly reviewed here in the context of the washout process. A more detailed treatment of the compartmental models is presented by Cobelli and Romanin-Jacur [12].
The simplest model which can be used to describe the washout process is the 'well-stirred compartment' model.
In this model, liquid entering a compartment of fixed volume is assumed to mix instantaneously with the fluid in the compartment. The concentration of analyte in the fluid leaving the compartment is the same as the concentration of analyte in the compartment. This process is shown schematically in Figure 6(a).
Let the dilution of analyte D be given by CCo, where C is the time varying analyte concentration and Co is the concentration of analyte injected into the system at time 0. If the concentration ofanalyte in the incoming liquid is zero, then the relationship between wash volume and dilution is: where Wis the volume ofliquid which has passed through the compartment since time 0, V is the volume of the compartment and A is the ratio of the concentration of analyte in the compartment at time 0 to the concentration of analyte injected.
A cascade of two well-stirred compartments, is shown in figure 6(b). In this model, liquid enters the first compartment, mixes completely, and then flows into the second compartment. Liquid mixes completely in the second compartment before exiting the system. This model is used to fit the data which resulted from experiments where the tip of the pipette was located close to the top of the column of liquid in the wash station. In this case, the two 'compartments' appear to represent portions of the liquid in the wash station which mix poorly. If the initial concentration in the first and second compartments at time 0 are Co and 0, respectively, and the concentration of analyte in the liquid entering the first compartment is zero, then the resulting equation for the output of this.model is given by: where V1 is the volume of the larger of the cascaded compartments, V2 is the volume of the smaller compartment and A is a constant which is related to the rate of flow of wash solution and to the compartmental volumes.
A model which appears to fit much of the experimental data is the two parallel compartments model. The analyte contained in one compartment, which is isolated from the flow of liquid, is transported into the second compartment, through which liquid flows. This model is shown schematically, in Figure 6(c). Severns and Adams [13] provide a detailed discussion of the assumptions which are involved for geometries similar to those of a thin film on the inside of a tube. The equation which describes the output of this model is: D A e-W/v1 + B e-W/V where V and V are a function of the volumes of the compartments, the rate of flow and the transport coefficient between compartments, and A and B are dependent upon the initial conditions in the compartments.
Taylor 10 and 14] and Aris 15] have described models to predict the dispersion of analyte in liquid flowing through a tube. The form of the dispersion of analyte depends upon whether the flow is laminar or turbulent and on the relative magnitudes of convective transport and diffusional transport of analyte. If the flow is laminar and transport ofsolute by diffusion negligible in comparison to convective transport, then the equations which describes the process are: D =B(1 W/A) W/A< D 0 W/A > (4) where B is the ratio of the concentration of analyte in the test solution to the concentration of analyte in the compartment at time 0 and A is the cross-sectional area of the tube,..If diffusion across the radius of the tube occurs in a time similar to the characteristic time for axial convective transport, anc if the flow is constant (so that the wash volume is directly proportional to the time), then the equation describing the process is: f o e-x ax the constant A is related to the initial dilution of analyte, and B is related to the rate of flow through the tube and the diffusivity of the analyte. If the flow is constant and of sufficient magnitude that turbulence occurs, the equation which describes the process is" D A W e-( w-Wo) /Bw (6) where A and B are constants related to the dispersion of the solute in the tube and W0 is the amount of liquid necessary to transport 1/2 of the analyte from the initial point to the measuring point.

An International Conference on Analytical Chemistry and Atomic Spectroscopy
This conference is being organized by the Analytical Division of the Royal Society of Chemistry, in conjunction wltl the Spectroscopy Group of the Institute of Physics, and will be held at the University ofBristol from Sunday to Saturday, 20 July to 26 July 1986. Application forms are now available and may be obtained from: Miss P. E. Hutchinson, Analytical Division, Royal Society of Chemistry, Burlington House, London W1V OBN. Tel: O1 437 8656.