acute kidney injury, urine proteomics, SELDI, mass spectroscopy
Background: Urinalysis is an important component in the assessment of acute kidney injury (AKI). Proteonomics is a rapidly developing approach in the analysis of physiological states. Several techniques have been developed to screen for protein populations. In this regard SELDI-TOF is a technique based on mass spectroscopy that is being utilized in proteonomics research.
Methods:For this study, clean catch or catheterized urine was collected from normals (n=18) and patients referred to the renal service with AKI. Based upon urine and serum chemistries, clinical parameters, and microscopic urinalysis, the urines were separated into those consistent with prerenal azotemia (n=17) and acute tubular necrosis (ATN) (n=29). Initially, 5 samples each were chosen from the pre-renal and ATN who had no preexisting renal disease. Other etiologies of AKI were not included in this analysis. The urine specimens were diluted 1:5 and deposited onto an H4 ProteinChip array using 50% acetonitrile as the binding buffer. This system captured the greatest spectral range with the SELDI-TOF evaluation (compared to SAX, WCX2, IMAC, and NP1 ProteinChips). Low (250) and high (300) laser intensities were utilized to ionize and desorb the protein molecules; the spectra were collected in a positive ion mode and analyzed with Ciphergen Peaks software (v 3.0).
Results: Five peaks with the high laser power were identified as potential candidates to discriminate between AKI due to prerenal or ATN causes. Those urines from the prerenal subjects were associated with detectable masses at 22.6 and 44.8 kilodaltons (KD); whereas subjects with ATN were noted to have urine with substantial masses at 11, 11.7, and 14.6 KD. The intensity of these peaks were then added together and normalized with the individual components of the discriminate peaks representing a percentage of the total. The prerenal and ATN subjects were then randomized in a training set consisting of 23 subjects and a testing set consisting of 23 subjects. Multiple linear regression was performed on the training set, and this allowed for 65% accuracy when applied to the testing set. Feed forward neural networks with hidden neuron layers ranging from 2-10 achieved similar predictive capability on the training set and testing sets.
Conclusions: Although the SELDI-TOF methodology may be a useful adjunct in the assessment of AKI and renal disease, we suggest that larger training sets will be necessary to effectively exploit this strategy.