Which biomarker would you like to discover today?

AB SCIEX, Darmstadt, Germany 

Acute kidney injury (AKI) is one of many clinical challenges where new biomarkers would be highly desirable to help improve patient care and clinical outcome. The KDIGO guidelines (Kidney Disease: Improving Global Outcomes) emphasise the importance of proper risk assessment in patients with potential AKI.(1) However, risk assessment is unsatisfactory if based on clinical assessment and currently available laboratory parameters alone, and imaging studies are also frequently unrevealing in this situation. 

Risk factors for AKI include hypovolaemia, for example, due to cardiovascular disease or surgery, inflammatory diseases, renal or liver disease, diabetes mellitus, radiologic contrast media and age >75 years.

Criteria for clinical assessment, such as RIFLE, allow stratification of AKI patients according to risk (Risk, Injury, Failure) and outcome (Loss, End Stage Renal Disease), based on the extent of decrease in the glomerular filtration rate (GFR) or the rise in serum creatinine level or the decrease in urinary output over prolonged duration.(2)

However, these criteria are not perfect.(3) First, serum creatinine may require 48h to rise; by this time 50% of kidney function might be lost. Second, it is well known that AKI can be oliguric (urine output <400ml/day) or non-oliguric (≥400ml/day).(4) Third, criteria such as 'rise in creatinine by 50%' require knowledge of baseline creatinine levels, which are often unavailable, because creatinine levels vary according to age, gender, ethnicity, muscle mass and even food intake. 

According to a report by the National Confidential Enquiry into Patient Outcome and Death (NCEPOD) in 2009, many deaths from AKI are still a result of delayed diagnosis, avoidable causes of AKI or inadequate risk assessment.(5) Recognition and risk assessment of AKI is important. First, the presence of AKI was shown to be an indicator of prolonged hospitalisation and increased mortality.(6) Second, RIFLE criteria were shown to correlate with mortality. For instance, mortality rates of 5%, 13% and 26% after cardiac surgery were reported for patients in RIFLE stages R, I and F respectively, with costs more than doubling from stage R to F.(7)

AKI is often classified into pre-renal, renal and post-renal, according to the underlying cause. Even in pre-renal AKI, there is evidence that inflammatory processes and possibly nitric oxide may play a role in the pathogenesis.(8) Depending on the cause of AKI, therapy includes the correction of hypovolaemia and administration of vasopressors and inotropes, removal of nephrotoxic drugs, treatment of sepsis (if present), and eventually renal replacement therapy. However, as mentioned above, in order to initiate timely therapy, an early diagnosis would be required. Such an early and improved diagnosis may constitute the basis for trials evaluating more effective therapy in the future.

We studied patients with AKI associated with sepsis. This is a particularly important setting, as mortality in sepsis was shown to increase from approximately 20% to >50% if both conditions are present.(9) Samples were also stratified according to the severity of renal injury as indicated by a post-hoc analysis of the maximum RIFLE stage. In general, it is advantageous to conduct biomarker studies in well-defined patient populations, otherwise clinical entities such as AKI might comprise a too-heterogeneous group of conditions, which would make the detection of common hallmarks difficult. 

The aim of the case study was to determine the feasibility of establishing a suitable method for the quantification of proteins in patients’ urine samples. The method should be simple, reliable and rapid because any biomarker study of sufficient power requires the analysis of samples from a large number of patients. We believe that the main reasons for previous failures in biomarker discovery are inadequate study design, lack of proper clinical classification of patients, failure to establish time courses and, most of all, inadequate number of analysed samples.

Urine samples from patients admitted to the intensive care unit (ICU) at the General Hospital of Vienna with a diagnosis of sepsis were collected, centrifuged to remove debris, frozen and stored at –80 degrees. The study was approved by the ethics committee of the Medical University of Vienna and informed consent was obtained from patients. Patients were classified post-hoc according to the maximum RIFLE stage (R, I, F or no AKI = 0) that was reached during the course of the stay in the ICU. Urine samples were selected that were obtained between eight hours after and 15 hours before the maximum RIFLE stage. The average (+/- standard deviation) time from sample collection to maximum AKI stage was 8 (+/-8) hours.

All selected samples were collected at enrolment, except for one sample collected 45 hours after enrolment from a subject that progressed to RIFLE F more than two days after enrolment. The protein concentration was determined, and samples from the same stage were pooled by mixing an equal amount of protein to obtain four pools designated 0, R, I, F with equal amount of total protein. The buffer was exchanged to an ammonium buffer containing urea by centrifugation through 10k membranes, followed by sample preparation according to the FASP protocol.(10) 

After reduction and alkylation of disulfide bonds, the samples were digested with trypsin at 37°C over night. A 1µg sample was separated on 50cm x 75µm ID columns (packed with 2µm C18 material) over a 90-min gradient at approximately 600bar. Each pool was measured once on an AB SCIEX TripleTOF® 5600+ mass spectrometer in information-dependent acquisition mode (IDA) for identification, and the four combined data files were searched against the human Uniprot database supplemented with common contaminant sequences using ProteinPilotTM version 4.5-b software.

Subsequently, each pool was measured in triplicate in SWATHTM mode for quantification, using the same sample amount and LC settings as for identification.(11) Settings for IDA analyses were: MS1 400–1250 m/z, 250ms; MS2: top20, intensity > 400, 80ms, Q1: unit resolution, CES 15, rolling collision energy, 200–1600 m/z, dynamic exclusion for 8s, all spectra recorded in high-resolution mode. SWATHTM settings were as follows: MS1 400–1000 m/z, 100ms; 20 SWATHTM experiments using 30Da isolation width, 90ms, 400–1500 m/z, CES 15.

SWATHTM involved fragmentation of all peptide precursor ions in each of the 30 Da isolations windows, which together covered the mass range from 400–1000 m/z. Subsequently, the fragment ion intensity was integrated over the elution profile of each peptide, based on the observed retention time and the fragment ions identified in the preceding IDA experiments. A false discovery rate analysis was performed on the list of proteins reported by ProteinPilotTM, and 626 proteins were found to comply with a global fitted FDR <1%.

Quantification was performed in PeakView® version 1.2, using the following settings: 25ppm ion library tolerance, import 626 proteins, integration window +/-2min and 100ppm, maximum five transitions for each peptide, and maximum five peptides per protein. A ‘transition’ is a specific combination of the m/z value of a peptide and the m/z of one of its fragment ions. Integration of the ion current associated with a transition over the elution time of a peptide, and summing up across all the transitions and all the peptides assigned to a protein yields a quantitative measure for each protein. 

Only peptides with a confidence above 95% ‘unique’ for a protein were used for quantification. For PCA analyses, samples were normalised according to Total Area Sums, and PCA was calculated using Pareto scaling in MarkerViewTM version 1.2.1.

T-tests for two-sample comparisons were calculated using Perseus 1.3.0.4 (MPI Martinsried) with S0 set to unit and p<0.05, and permutation-based correction for multiple hypothesis testing to detect significant regulation. Microsoft Excel was used for plots.    

High reproducibility of the number of proteins that can be quantified is another key merit of the data-independent SWATHTM technique. Even on the most basic level, that is, transitions (observation of fragment ion areas), quantitative data were available for more than 70% of data points in all three out of three technical repeats (Figure 1A). Even if one transition failed to be observed, a peptide could still be quantified by the other transitions associated with the peptide, and analogous considerations held true for peptides associated with a protein. For this reason, the percentage of peptides and proteins that were quantified in all three runs was even higher, approximately 90%.

These values constitute remarkable figures of merit, particularly in comparison with quantitative studies employing data-dependent acquisition schemes, where the overlap on the peptide level between technical repeats is often only approximately 35% to 60%.(12) Using data-independent SWATHTM acquisition, the percentage of proteins quantified at least three times could be further enhanced by performing four or even five measurements, which might possibly lead to an almost complete set of proteins with three or more data points.

This might prove helpful for downstream statistical analysis, where missing data points constitute a notorious cause of inconsistency. Notably, the median coefficient of variation (CV) of technical replicates was around 10% (Figure 1B), which compares favourably with 18% median CV reported previously for urine analysis using data-dependent acquisition.(13) The coefficient of variation (CV) was below 20% for 80% of the data in our preliminary experiments, and below 30% for 90% of the data. We consider this acceptable for label-free quantification, in particular as we observed that the biological variance of proteins in human urine seemed to be considerably higher.

The number of samples N required for a study to reach a statistical power of, for instance, 80% or higher (that is, the probability to detect a finding based on the given significance level, for example, alpha error <5%) depends on the relationship between the effect size and the square root of the variance. As long as the biological variance is high compared to the experimental variance, the number of samples N that has to be analysed depends mainly on the biology (effect size and variance). Importantly, the quality and reliability of SWATHTM quantification can be monitored in terms of the CV% of the transitions that are associated with differentially regulated peptides and proteins. We could show that exceptional CVs of <5% or 10% could be achieved for a large fraction of the most intense transitions (Figure 1C).

We identified 626 proteins in the four 90-min gradients at 1% FDR, ensuring that 99% of protein identifications are correct and reliable. As the human ‘urinary core proteome’ was reported to comprise around 500 proteins,13 we consider the sensitivity of our analytical method as adequate and sufficient for biomarker discovery in urine. 

A principal component analysis (PCA) of the 12 SWATHTM measurements (four samples x three technical replicates) clearly separated the four RIFLE groups F, I, R and 0, that is, no AKI (Figure 2A). Principal Component Variable Grouping (PCVG) using three principal components, an angle of 45 degrees, and a minimum distance from origin of 0.1 permitted the classification of proteins into groups (Figure 2B).

Figure 2C illustrates the profile of the response of three proteins that display upregulation in RIFLE F and which might therefore be indicators of the development of RIFLE stage F (failure). Figure 2D illustrates the extracted ion chromatograms (XICs) of one ‘transition’ measured across 12 samples, reflecting the signal intensity of the respective fragment ion over the elution profile of the peptide. The example illustrates upregulation in RIFLE stage I, and even more so in stage F.

Many other proteins also displayed interesting patterns. Because of the fact that three technical repeats were acquired for each sample, t-tests, which gave a p value for each protein, could be calculated, for instance for a two-sample comparison of RIFLE stages F and 0 (no AKI). After log-transformation and plotting of the p values versus the regulatory ratios (‘volcano plot’), interesting candidate proteins could be identified by low p-values and high regulatory ratios (Figure 3). Both upregulation and downregulation in RIFLE stage F might indicate useful markers.

Of note, we were able to observe and confirm regulation of known and previously published possible biomarkers of AKI, such as NGAL (neutrophil gelatinase-associated lipocalin).(14) This indicates that the technology is ready for an analysis of larger numbers of samples. Of course, an actual clinical study will require analysis of samples from individual patients and from larger numbers of patients. Such analyses are currently underway. The high acquisition speed of the instrument permits analysis of a large number of samples within a reasonable time frame, which contributes to making studies with adequate statistical power feasible.

We conclude that data-independent SWATHTM analyses of urinary samples are simple, rapid and sensitive, and reliable enough to permit the analysis of large numbers of samples for adequately powered studies aimed at discovering early biomarkers indicative of AKI. As the technology can also be applied to all other kinds of samples including body fluids or tissue, we expect that MS/MSALL with SWATHTM Acquisition will become a valuable tool in clinical discovery.

1. KDIGO Clinical Practice Guideline for acute kidney injury. Kidney Int Suppl 2012;2(1):1–138.
2. Bellomo R et al. Acute Dialysis Quality Initiative workgroup. Acute renal failure – definition, outcome measures, animal models, fluid therapy and information technology needs: the Second International Consensus Conference of the Acute Dialysis Quality Initiative (ADQI) Group. Crit Care 2004;8(4):R204–12.
3. Winterberg PD, Lu CY. Acute kidney injury: the beginning of the end of the dark ages. Am J Med Sci 2012;344(4):318–25.
4. Thadhani R, Pascual M, Bonventre JV. Acute renal failure. N Engl J Med 1996;334(22):1448–60.
5. National Confidential Enquiry into Patient Outcome and Death. Adding insult to injury. NCEPOD 2009.
6. Barrantes F et al. Acute kidney injury criteria predict outcomes of critically ill patients. Crit Care Med 2008;36(5):1397–403.
7. Dasta JF et al. Costs and outcomes of acute kidney injury (AKI) following cardiac surgery. Nephrol Dial Transplant 2008;23(6):1970–4.
8. Bonventre JV, Zuk A. Ischemic acute renal failure: an inflammatory disease? Kidney Int 2004;66(2):480–5.
9. Hoste EA et al. Acute renal failure in patients with sepsis in a surgical ICU: predictive factors, incidence, comorbidity, and outcome. J Am Soc Nephrol 2003;14(4):1022–30.
10. Wiśniewski JR et al. Universal sample preparation method for proteome analysis. Nat Methods 2009;6(5):359–62.
11. Gillet LC et al. Targeted data extraction of the MS/MS spectra generated by data-independent acquisition: a new concept for consistent and accurate proteome analysis. Mol Cell Proteomics 2012;11(6):O111.016717.
12. Tabb DL et al. Repeatability and reproducibility in proteomic identifications by liquid chromatography-tandem mass spectrometry. J Proteome Res 2010;9(2):761–76.
13. Nagaraj N, Mann M. Quantitative analysis of the intra- and inter-individual variability of the normal urinary proteome. J Proteome Re. 2011;10(2):637–45.
14. Devarajan P. Neutrophil gelatinase-associated lipocalin: a promising biomarker for human acute kidney injury. Biomark Med 2010;4(2):265–80.