Multi-region proteome analysis quantifies spatial heterogeneity of prostate tissue biomarkers

Abstract

It remains unclear to what extent tumor heterogeneity impacts on protein biomarker discovery. Here, we quantified proteome intra-tissue heterogeneity (ITH) based on a multi-region analysis of prostate tissues using pressure cycling technology and SWATH mass spectrometry. We quantified 6,873 proteins and analyzed the ITH of 3,700 proteins. The level of ITH varied depending on proteins and tissue types. Benign tissues exhibited more complex ITH patterns than malignant tissues. Spatial variability of ten prostate biomarkers was validated by immunohistochemistry in an independent cohort (n=83) using tissue microarrays. PSA was preferentially variable in benign prostatic hyperplasia, while GDF15 substantially varied in prostate adenocarcinomas. Further, we found that DNA repair pathways exhibited a high degree of variability in tumorous tissues, which may contribute to the genetic heterogeneity of tumors. This study conceptually adds a new perspective to protein biomarker discovery: it suggests that recent technological progress should be exploited to quantify and account for spatial proteome variation to complement biomarker identification and utilization.

Figure 1.. Study design.

(A) Hematoxylin and eosin staining of the fresh frozen prostate tissue from three individuals who have contributed to BPH (non-tumorous) and matching acinar or ductal ADCA. Green, orange, and blue lines depict regions diagnosed by a pathologist as BPH, acinar, and ductal tumors, respectively. Black circles indicate where the punches were made. (B) Overall measured variation of protein expression was partitioned into biological and technical variation including inter-patient variation, inter-tissue variation, intra-tissue variation, and technical variation from MS analysis and batch variation. Three or six punches were sampled from each tissue type, followed by PCT-SWATH analyses in technical duplicate. The samples were shuffled and analyzed in 10 batches of six samples.

Figure S1.. Benign and malignant prostate tissue from three individuals.

Hematoxylin and eosin staining of the fresh frozen prostate tissue used in this study. Amplified views of representative region in each area were shown in (B–I) as indicated.

Figure S2.. Unsupervised clustering of 3,700 proteins quantified with at least two concordant peptides.

Figure 2.. Consistency of technical and total variance.

(A) Correlation of technical variances estimated independently for different samples. Technical variance is estimated from technical replicates. (B) Correlation of total variances (between punches) estimated independently from punches from different tissue samples (different patients, different tissue types).

Figure S3.. Dependence of technical variance on protein intensity.

Proteins are divided into eight bins with roughly the same number. X-axis shows the mean intensity value of each bin, and Y-axis shows the log10 technical variance.

Figure S4.. Density curves of biological variance.

Occasionally, our estimate of the technical variance was larger than the variation between punches, after technical replicates were averaged per punch. This resulted in a negative estimate of the biological variance, which is of course infeasible. We assumed that those proteins have a biological variance close to zero; thus, the total variance is mostly reflecting technical variance. Therefore, we used the distribution of negative scores as a background distribution (null distribution) for the null hypothesis that there is no biological variance between punches. The blue curve shows the negative part of the distribution mirrored on the positive side. The distribution of observed biological variance estimates (red) is clearly above that background distribution.

Figure 3.. Correlation of biological variance between patients and tissue types.

Each dot represents one protein. (A) Distributions of biological variance estimates. Inter-patient variances and inter-tissue variances are based on averaging the measurements of at least three punches. Intra-tissue variance was first determined independently per patient and tissue type, and then averaged. (B) Biological variance between tissue of the same patient versus variance between punches of the same patient and tissue. (C) Biological variance between different patients but same tissue type versus variance between punches of the same patient and tissue. (D) Biological variance between the same tissue types in different patients versus variance between different tissue types of the same patient.

Figure 4.. Intra-tissue heterogeneity in tumorous and non-tumorous tissue.

(A) Biological variance among punches from the same tissue region was considered as the degree of intra-tissue heterogeneity for the respective tissue type. Degree of intra-tissue heterogeneity for each protein in benign versus malignant tissue are shown and colored according to classification. (B) GO enrichment analysis of four protein categories from (A). Length of horizontal bars indicates the significance of the enrichment. (C) Intra-tissue heterogeneity of biochemical pathways. Each triangle is the average biological variance (intra-tissue heterogeneity) of all quantified proteins from the respective pathway. Degree of intra-tissue heterogeneity for each pathway in benign versus malignant tissue is shown. Pathways were grouped according to their variability in benign and malignant tissue.

Figure 5.. Immunohistochemical validation of representative proteins.

The top proteins from four ITH groups in BPH and malignant (ADCA) prostate tissue were validated using a TMA with two representative tissue spots of each patient.

Figure S5.. Staining images of protein expression for the TMA.

Six proteins (ACPP, ABCF1, NUP93, CUTA, CRAT, and FSTL1) were measured using IHC in a TMA containing tissue samples from 83 patients from an independent cohort, including 35 patients with BPH and 48 patients with prostate ADCA.

Figure 6.. Correlation between mass spectrometry–based (MS) variance estimates and TMA homogeneity.

(A) shows benign tissues whereas (B) depicts tumor tissues. The concentrations of CRAT and NUP93 were almost zero in the benign tissue samples. Thus, it is virtually impossible to estimate their intra-tissue variation in benign tissues. The correlation between MS-based variance and TMA homogeneity was, however, computed without excluding these two proteins. NUP93 was slightly off the regression curve because its signal in IHC was relatively weak.