Selective degradation of serum proteins is likely responsible for the spurious differences in innate immunity proteins observed in a type 1 diabetes study

Introduction

Type 1 diabetes (T1D) is an autoimmune disease that occurs when insulin-producing beta cells in the pancreas are destroyed by the body's immune system. Efforts are being made to develop biomarkers that could predict the risk of developing T1D or complications associated with the disease¹. Since pathological tissues from patients with T1D are very difficult to obtain for biomarker studies due to ethical and practical issues, proteomic analyses of serum proteins may provide useful biomarkers relevant to diseases because serum proteins may be secreted by cells at the pathological sites². Comprehensive analysis of the serum proteome is a challenging task using current proteomic technologies due to its extraordinary complexity and high dynamic range in concentration^3,4. Proteomic analyses are further complicated by the potentially large variation between individuals and relatively small differences between study groups. Furthermore, serum protein levels may be influenced by a large number of parameters such as subject age, sex, geographic or ethnic origin, environmental factors and disease states. Therefore, the quantification of serum proteome requires well designed strategies that incorporate different technologies and experimental designs at different phases, each addressing a different set of problems.

In a 2013 issue of The Journal of Experimental Medicine, Qibin Zhang and colleagues⁵ described the discovery, verification and validation of several serum proteins/peptides that were drastically different between type 1 diabetes (T1D) patients and healthy controls. The authors used label-free quantitative LC-MS-based proteomics to discover 24 proteins that showed significant changes between T1D and healthy control samples and further verified these differences using a MRM-MS based multiplexed peptide assays with spiked stable isotope-labeled standard (SIS) for normalization purpose. It would be of great scientific and clinical importance and interest if these biomarker candidates were to be further validated in independent samples and in different laboratories. Therefore, we attempted to replicate the results by using the same methods to analyze a large panel of 145 T1D patients and 156 autoantibody negative (AbN) control subjects (PANDA data set) collected in the same geographical area, during the same period of time and by the same investigators. Unfortunately, our results do not support the major findings in the report.

Materials and methods

Results and discussion

A total of four plates of serum samples from our PANDA sample cohort (145 T1D patients and 156 AbN control subjects) were analyzed using MRM assays for twelve peptides that have good MS signals (s/n>5), including NIQSLEVIGK, LLDSLPSDTR, FQPTLLTLPR, TNLESILSYPK, LVLLNAIYLSAK, ITQVLHFTK, TLEAQLTPR, ELDESLQVAER, AGALNSNDAFVLK, TFTLLDPK, DIPTNSPELEETLTHTITK and TGAQELLR. As displayed in Figure 1, five of the twelve peptides (LLDSLPSDTR, FQPTLLTLPR, TNLESILSYPK, LVLLNAIYLSAK and ITQVLHFTK) whose serum levels are significantly different in the DASP verification and/or blind sample sets are not significant at all in our large PANDA sample set (p>0.05). TGAQELLR showed marginal significance (p=0.03). Although the remaining 6 peptides (NIQSLEVIGK, TLEAQLTPR, ELDESLQVAER, AGALNSNDAFVLK, TFTLLDPK and DIPTNSPELEETLTHTITK) are significantly different between the T1D and control group in our PANDA sample set, the inter-group differences as measured by fold change (FC) are very small (FC = 1.0±0.1). These results are in sharp contrast to the dramatically altered concentration for several peptides observed in the T1D patients in the initial report, including NIQSLEVIGK (FC of -1.1 in PANDA versus 8.5 in DASP_V), FQPTLLTLPR (FC of 1.0 versus 2.6), TNLESILSYPK (FC of 1.0 versus 2.2), and LLDSLPSDTR (FC of 1.1 versus 2.1).

Figure 1. Relative abundance for 12 peptides.

Dot plots showing the peptide abundance in each study subjects. Three sample sets (PANDA, DASP_V and DASP_B) were analyzed in our laboratory using identical procedures. The black horizontal line in each group represents the mean value for the group. Relative peptide abundance (y axis) was calculated as the ratio between the endogenous peptide and its spiked-in, heavy isotope–labeled synthetic peptide. The numbers in the parentheses are the subject numbers in each group. Red double-asterisk and single-asterisk denote a p value <0.01 and <0.05, respectively.

In order to exclude the possibility of different MRM assay platforms causing the discrepancies between the published and our results, we obtained the samples used in the initial report (kindly provided by the authors). The samples include part of their DASP verification cohort (DASP_V, 99 controls and 25 T1D patients) and their entire DASP blind (DASP_B) validation cohort (10 controls and 10 T1D patients). We processed these samples and performed MRM assay using the same methods used to analyze our PANDA samples. The dot plots for DASP sample cohorts are also shown in Figure 1 (DASP_V and DASP_B) together with the ones for our PANDA samples. The corresponding FC and p values are listed in Table 1. The FC and p values from our measurement in the DASP samples correlated very nicely with the authors’ published results, clearly indicating that the techniques used in the two laboratories yield very similar results on the same sample sets.

We further compared the results for the two sample cohorts and found out that 6 of the 7 transitions/peptides whose serum levels are decreased in T1D groups of DASP_V and/or DASP_B sample sets are also significantly lower in the T1D group of our PANDA sample set (TGAQELLR, TLEAQLTPR, ELDESLQVAER, AGALNSNDAFVLK, TFTLLDPK and DIPTNSPELEETLTHTITK, corresponding to 5 different proteins). In contrast, the transitions/peptides whose serum levels are significantly increased in T1D group from the DASP sample set either showed an significantly opposite trend (FC=-1.1, p=4.92E-03 versus FC=6.8, p=2.07E-06 in DASP_V for NIQSLEVIGK, corresponding to protein PPBP) or did not change significantly in our PANDA sample set (p>0.05 for LLDSLPSDTR, FQPTLLTLPR, TNLESILSYPK and LVLLNAIYLSAK, corresponding to the sample protein SERPING1).

How can one explain the drastic discrepancies between these two studies? Comparison of the serum concentration distribution of the 5 reported significantly elevated transitions/peptides in PANDA versus DASP sample sets (Figure 1) indicates that their expression levels are similarly high in the patients (T1D) from both studies. In contrast, the PANDA controls have similarly high levels as the patients from PANDA and DASP, while the DASP controls have dichotomous distribution for these five transitions/peptides. A few of the DASP controls have comparable levels as observed in the patients and PANDA controls, with a large proportion of the DASP controls have undetectable levels of those peptides (Figure 1). Our results clearly indicate that the DASP control samples contain many outliers that are responsible for the reported differences between T1D and controls in the previous study¹. To identify factors that may account for the discrepancies between controls in the two studies, we examined various demographic parameters available to us. Sex is unlikely to explain the differences as the PANDA and DASP cohorts contain subjects of both genders. The only apparent demographic difference between the controls in the two studies is the age distribution. The mean ages are 30.1 and 20.7 for controls from the PANDA and DASP datasets, respectively. However, the peptide concentration does not significantly vary according to subject’s age in either the AbN controls or the T1D groups from PANDA (data not shown), an observation consistent with the authors’ conclusion. The significant fold changes for the proteins in the original report are clearly caused by the extreme low serum protein levels in the control group, rather than the increased level in the patient group, as shown by the distinct dichotomous distribution and undetectable levels in many DASP controls (Figure 1). This situation is most likely caused by degradation of proteins that are sensitive to sample collection and storage conditions, a phenomenon well known to protein biochemists^6,7. Interestingly, some proteins may be more susceptible to degradation than other proteins as suggested by the dichotomous distribution of the intensities of the 5 transitions derived from two proteins (PPBP and SERPING1) and the continuous distribution of the intensities of the 7 transitions derived from 6 other proteins (GSN, SERPIND1, C4A, CLU, PGLYRP2 and KNG1).

In conclusion, our results directly question the validity of the conclusions of the initial report and highlight the importance and necessity to conduct biomarker discovery and validation studies using well collected and preserved samples as well as sufficiently powered sample sets.

Data availability

F1000Research: Dataset 1. Peak areas of MRM transitions in PANDA and DASP sample set, 10.5256/f1000research.5384.d36443⁸