Impact of sampling technique on appraisal of pulsatile insulin secretion by deconvolution and cluster analysis

Little is known about the optimal experimental conditions for assessing pulsatile insulin secretion in vivo. To address this, we employed a recently validated canine model (n = 12) to determine the consequences of 1) sampling from the systemic circulation (SC) vs. the portal vein (PV), 2) sampling intensity and duration, and 3) deconvolution vs. cluster analysis on assessing pulsatile insulin secretion. PV vs. SC sampling resulted in a ~40% higher pulse frequency by deconvolution (9.0 ± 0.5 vs. 6.6 ± 0.9 pulses/h, P < 0.021 and cluster analysis (7.5 ± 0.3 vs. 5.6 ± 0.6 pulses/h, P < 0.01) due to a higher signal-to-noise ratio (19 ± 4.8 PV vs. 12 ± 1.8 SC). PV sampling also disclosed a higher calculated contribution of the pulsatile vs. nonpulsatile mode of delivery to total insulin secretion (57 ± 4 vs. 28 ± 5%, P < 0.001). Analysis of the relevance of sampling intensity revealed that 1-min data yielded a markedly higher estimate of pulse frequency with PV sampling than 2-min data (9.0 ± 0.5 vs. 5.4 ± 0.5, P < 0.02, deconvolution; 7.5 ± 0.3 vs. 4.3 ± 0.6 pulses/h, P < 0.001, cluster). Optimal sampling duration was shown to be 40 min or more. We conclude that the resolving power of the analytical tool, the anatomic site of blood withdrawal, the frequency of blood sampling, and the duration of the total observation interval all significantly influence estimated insulin secretory pulse frequency and the fraction of insulin secreted in pulses. With the assumption that PV 1-min insulin data constitute the 'gold standard,' our in vivo inferences of 7.5- 9.0 insulin pulses/h closely recapitulate in vitro islet secretory activity.