This study demonstrates that the consecutive transfer of CSF samples between tubes has significant impact on the measured concentration of Aβ42. Figure 2A reveals the potential for concentration levels in neat CSF pools to be artificially reduced from within the ‘normal’ (that is, CTRL) range to pathological levels in three tube transferals. In a different CSF pool, tested three times (in two different assay kits), this was the case in just one transfer (Figure 2B and C). This difference between transfer steps 0 to 1 is interesting. It is possible that the volume difference between transfer tube 0 and transfer tube 1 in the pilot study could have had an impact on this result. However, this is considered unlikely for reasons discussed below. Alternative suggestions could be that there was a fault with pilot tube 0 (given the close similarity between results in the six independent tubes used in the replication rounds it seems unlikely a fault lay there), an anomaly caused by human or detection error (once again it seems more reasonable to suspect the pilot result), or a difference in the pool matrices which may have altered the behaviour of the analyte in the conditions of the first transfer step. When the Aβ42 results for the neat AD and CTRL pools of the pilot study and the replication study (single-plex and triplex) are compared over the next three transfer steps (Figure 2) a linear trend can be observed between steps 2 to 4. Furthermore, this trend was demonstrated to continue over steps 5 to 7 in the replication rounds. Table 1 demonstrates very comparable linear regression results, with strongly overlapping confidence intervals, for all Aβ42 assays conducted.
The addition of the non-ionic surfactant Tween 20 to pooled samples prior to aliquoting had a mitigating effect on the reduction in measured Aβ42 over the transfer series, although reduction was still significant. The storage concentration of Tween 20 used in this study was 0.05% and after 1:8 sample dilution would have been approximately 0.006%. The generally accepted critical micelle concentration (CMC) for Tween 20 is 0.007%, but micelle formation has been shown to initiate at 0.002% . This suggests that many of the Tween 20 molecules in our samples would still be expected to be in a micelle arrangement during the assay  and, as such, be in competition for tube surface and liquid/air interface distribution with other hydrophobic molecules. It has been shown that Tween 20 may also prevent oligomerisation of Aβ42  and this may apply to aggregation more generally. Our study does not elucidate the relative involvement of the two potential mechanisms. Figure 2 and Table 1 show that treating samples with Tween 20 significantly reduced concentration loss per transfer step, relative to their neat counterpart. However, it is worth noting that a dramatic decrease in concentration was observed between transfer steps 5 to 7 in the TCT pool of the replication study. This would be consistent with a decrease in the concentration of Tween 20 molecules (either by complete loss of all Tween 20 molecules or concentration falling far enough below the CMC that micelle numbers no longer formed effective surface competition) sufficient to allow Aβ42 adsorption comparable to that in neat CSF. This effect was not observed in any other pool, but consistent replication (CV% at Transfer 6 = 3.1%, at Transfer 7 = 19.0%) suggests it is unlikely to be an error artefact.
Data show that the consecutive transfer of CSF samples between tubes has a significant impact on the measured concentration of Aβ38. Figure 3 and Table 1 show a strong linear tendency for concentration reduction in neat CSF. This effect was greatly mitigated in the same samples treated with Tween 20. Over four transfer steps reduction did not reach significance in Tween treated samples, but over all seven steps significant reduction was observed.
This study demonstrates that the consecutive transfer of CSF samples between tubes has a significant impact on the measured concentration of Aβ40. Figure 3 and Table 1 show a strong linear tendency for concentration reduction in neat CSF. This effect was greatly mitigated in the same samples treated with Tween 20. As with Aβ38, over four transfer steps reduction did not reach significance in Tween treated samples, but over all seven steps significant reduction was observed. It is interesting to note that the starting concentration of Aβ40 was nearly twice that of Aβ38 and, accordingly, linear regression is calculated to follow the same relationship. This demonstrates the apparent concentration dependency of transfer loss rather acutely and is a trend observed in all other analytes.
Data show that consecutive transfer of CSF samples between tubes had a much smaller effect on T-tau concentrations than on the Aβ peptides. The results of the pilot showed a non-significant trend for a reduction in concentrations of T-tau in neat CSF and no evidence of a reduction following the addition of Tween 20. However, the P values for neat AD (P = 0.06) and control (P = 0.066) pools approached significance. Due to the difference in dilution factor between the pilot (1:2) and the replication study (1:4), and given the apparent tendency for protein loss to be concentration dependent, results may not be directly comparable between the two studies. The replication study showed significant reduction in T-tau concentration over seven transfer steps in both neat pools (Figure 4, Table 1). The NAD pool of the replication study reached significant reduction over the initial four transfer steps, thus providing a case that T-tau can be affected within this number of transfers. Tween 20 CSF pools remained unaffected even over the seven transfer steps. Compared with Aβ peptides, neat and Tween 20 T-tau results show a proportionally lower rate of concentration reduction per transfer step (Table 1). Transferring CSF between multiple surfaces could, therefore, create an artificially low Aβ42 to T-tau ratio and risk false positive diagnosis of AD in patients and research participants.
The transfer effects we observed – principally for Aβ peptides, but also, to a lesser extent, for T-tau measurement – may well have significant influences in practice. Not only individual levels of these analytes, but also the ratio of T-tau to Aβ42 are used in clinical and research criteria contributing to diagnosis of AD . Additionally, it is possible that Aβ peptide ratios could be altered over a number of transfers, given the smaller percentage decrease in peptides 38 and 40 relative to 42, or by the potential for non-linear reduction, despite a common linear tendency.
In a previous study , we identified sample volume as a potential confound to the measurement of Aβ42 concentration but not T-tau. There were inter-study volume differences between the pilot and the replication studies and intra-study volume differences in the pilot alone.
In the pilot, although the volume and, therefore, the surface area, of each sample was equivalent in storage (925 μL), it was not possible to maintain consistent 200 μL volume between all the transferral stage aliquots. Based on the previous data (0.8 pg/mL increase per 10 μL increase in control CSF Aβ42, 0.74 pg/mL increase per 10 μL increase in AD CSF Aβ42) the following discrepancies would be expected relative to an aliquot at 200 μL:
Aβ42 Tube 0 (215 μL):
Aβ42 Tube 4 (110 μL):
Therefore, volume effects are not likely to have been significantly above noise and thus sufficient to bias the results of this study.
In the replication study storage aliquots were 1000 μL and transfer aliquots were kept equal at 100 μL so as to exclude any effect on volume. It should be noted that the samples used in the pilot and replication studies were of different pools, and the comparisons of this study are based on proportional concentration loss not absolute values. Direct comparisons of Aβ42 between the pilot and replication study are not likely to be valid.
Murphy et al.  have identified that short term storage of CSF samples on dry ice can lower the sample pH through intrusion of CO2. pH change can affect the ‘tertiary and quaternary structure, enzymatic rate constants, solubility, tendency to aggregate, susceptibility to chemical degradation and propensity to adsorb to surfaces’ of constituent proteins . In our study the original, large volume samples were transported between the Sahlgrenska and London laboratories on dry ice. Once in London they were thawed, divided into aliquots and not subsequently exposed to dry ice. Our results are not, therefore, attributable to this effect.
Standard procedure for CSF collection by lumbar puncture  (However, it should be noted that touching any part other than the plastic head of the needle may raise sterilisation issues) involves the fluid being passed typically through a 20- to 22-gauge spinal needle (sometimes with a catheter) and dripped into a collection tube. In some cases, and in some centres, CSF is aspirated using a syringe. Baseline and diagnostic tests may then be run on the sample, before it is aliquoted into smaller volume storage tubes. Thus, CSF can encounter two or three different containers before reaching clinical diagnostics, and three or four or more before reaching storage for research or re-testing. Figures 2, 3 and 4 collectively show that this could lead to compromised diagnoses, misleading research data and discrepancy between results. Further attention needs to be directed toward how every step of collection may compromise the accuracy of current assays relative to in vivo reality. A number of these issues have been addressed in other studies [3, 14].