A label-free approach by infrared spectroscopic imaging for interrogating the biochemistry of diabetic nephropathy progression

Abstract

Routine histology, the current gold standard, involves staining for specific biomolecules. However, untapped biochemical information in tissue can be gathered using biochemical imaging. Infrared spectroscopy is an emerging modality that allows label-free chemical imaging to derive biochemical information (such as protein, lipids, DNA, collagen) from tissues. Here we employed this technology in order to better predict the development of diabetic nephropathy. Using human primary kidney biopsies or nephrectomies, we obtained tissue from four histologically normal kidneys, four histologically normal kidneys from diabetic subjects and five kidneys with evidence of diabetic nephropathy. A biochemical signature of diabetic nephropathy was derived that enabled prediction of nephropathy based on the ratio of only two spectral frequencies. Nonetheless, using the entire spectrum of biochemical information, we were able to detect renal disease with near perfect accuracy. Additionally, study of sequential protocol biopsies from three transplanted kidneys showed biochemical changes even prior to clinical manifestation of diabetic nephropathy. Thus, infrared imaging can identify critical biochemical alterations that precede morphological changes, potentially allowing for earlier intervention.

INTRODUCTION

Kidney transplantation is often an effective treatment for end-stage renal disease (ESRD) (1, 2); however, transplant failure is common. In 2012, 28% of those receiving a kidney transplant had diabetic nephropathy as the primary cause of ESRD (3). Unfortunately, this group also has the lowest five year graft survival compared to ESRD from hypertension, cystic disease, glomerulonephritis, or from other causes (3). Allograft function is monitored using serum creatinine to monitor for complications but unfortunately, they are usually only detected after an irreversible decline in renal function. Some institutions may use surveillance biopsies performed over the first year for early identification of subclinical complications (4, 5). Histological evaluation of renal biopsies by a pathologist remains the gold standard for obtaining critical diagnostic and prognostic information after transplantation (6), and has been shown to impact subsequent treatment (7). However, early histological evidence of diabetic nephropathy including glomerular basement membrane and mesangial thickening are not seen until two or more years after transplant, thus highlighting the lack of advanced tools that can be used for the prediction of graft outcome (6).

Recent dramatic advances have transformed infrared (IR) imaging from a research tool into a potentially powerful clinical tool, providing insight in a label-free, non-perturbing manner into the biochemistry of tissue (8–10). IR imaging represents a potentially powerful adjunct to current pathology techniques due to it’s ability to provide complementary biochemical information that would otherwise not be accessible using conventional staining approaches. IR spectroscopy is based on the principle that different regions of mid-IR light are quantitatively absorbed by different biomolecules present within tissues such as proteins, lipids, DNA, RNA, collagen, glycogen, and carbohydrates (11). Different tissue biomolecules, have different characteristic IR absorption spectra. The principal for much of the work using IR imaging for tissue pathology has focused on examining these IR spectra, also termed spectral fingerprints, and using these fingerprints to predict cell type/disease status or replicate staining patterns with a high level of accuracy (12, 13). Image resolutions approaching one micron are now available permitting visualization of key structures in renal biopsies and subsequent extraction of biochemical information (Fig 1) (8, 14). In visible microscopy, an image typically has three channels for every pixel, typically a red, a blue and a green channel. In IR imaging, every single pixel within the image comprises of an entire IR spectrum typically comprised of hundreds of biochemical channels (Fig 1). IR imaging can potentially provide additional diagnostic/prognostic information in renal biopsies of value for treatment and prognosis. Fourier transform infrared spectroscopy is routinely applied to the analysis of renal stones and it has recently been used for the detection of 2,8-dihydroxyadenine crystals located in renal tubular lumen (15). Previous work has demonstrated that spectroscopic differentiation of pathological conditions can be achieved in the absence of visual clues (16, 17).

Figure 1.

Schematic of a typical IR imaging system. An IR system has both visible and IR light source/detection systems. (A) The visible light is used to visualize the sample and find the area of interest on the tissue section. The CCD visible detector in the system collects the visible light and creates a data cube with an RGB image (3 bands, Z) coupled with spatial dimension (X,Y). After the region is found, the mirrors in the system flip to the IR source/detector, and an IR source is used to obtain an image. (B) As mid-IR passes through the sample, different regions are absorbed by different biomolecules (glycogen, DNA, proteins, etc.) to give a biochemical fingerprint of the tissue, which is collected using a detector sensitive to IR. This creates an image with possibly more than 200 bands of data/images (Z). This also is coupled with spatial data (X,Y) for each spectral band recorded in the Z-axis.

RESULTS

We first identified the signature of isolated diabetic nephropathy in native kidney tissue. Periodic Acid Schiff (PAS) stained kidney biopsies and nephrectomy tissue was grouped into three categories: histologically-normal non-diabetic (NL, n=4), histologically-normal diabetic (NLD, n=4) and diabetic with evidence of diabetic nephropathy (DN, n=5) (patient details in table 1 supplemental data). PAS staining is primarily used to stain carbohydrate macromolecules in renal biopsies. Formalin-fixed paraffin-embedded tissues were serially sectioned onto a glass slide (for PAS staining) and a barium fluoride slide (for IR imaging). High-resolution IR images were acquired and IR spectra extracted from the glomerular basement membrane (GBM), tubular basement membrane (TBM) and mesangium (M). An average IR spectrum was derived from up to six glomerular or tubular cross sections of each specimen (Fig 2). In all three glomerular components, a pronounced difference was particularly found in the 1120–1000cm⁻¹ region from the entire mid-IR spectral range (3850–900cm⁻¹, Fig 2). Increases are seen in the glycosylation (1030cm⁻¹) and the DNA and glycosylation associated (1080cm⁻¹) peaks in the GBM, TBM and M of the DN biopsies. Importantly, the NL and NLD cohorts showed very similar spectra. Thus the diabetic signature we were detecting was not due to the patient’s diabetic status but due to diabetic nephropathy.

Figure 2.

Spectral data was extracted from glomerular basement membrane (GBM), tubular basement membrane (TBM) and mesangium (M) from IR images scans. The average spectra for diabetic nephropathy (DN), normal diabetic (NLD) and normal non-diabetic (NL) groups is shown in the top row (A–C), while individual spectra are shown below (D–F). The DN groups showed increased in the glycosylation (1030cm⁻¹) and DNA and glycosylation associated (1080cm⁻¹) peaks in all three structures compare to the control groups (NLD and NL).

One feature of diabetic nephropathy is the increase in glomerular mesangial fraction of surface area (MFSA) (18). MFSA was computed from the PAS stained sections for up to 6 glomeruli per patient (Fig 3A). The IR spectrum of the mesangium from adjacent tissue sections of the same glomeruli (approximately 350 pixels/spectra per glomeruli) were extracted and a simple spectral ratio determined (Fig 3B). Biopsies with diabetic nephropathy had markedly increased MFSA and spectral ratio of 1030 cm to 1080 cm (Figs 3A and 3B) with MFSA. A small earlier pilot study (data not shown) found that Kimmelstiel-Wilson nodules of the mesangium had an increase in the 1030:1080 ratio. A plot of the 1030:1080 ratio against the MFSA showed that the DN biopsies (high MFSA and high spectral ratio), and the histologically normal biopsies (NL and NLD, low MFSA and low spectral ratio) formed distinct clusters (Fig 3C). Interestingly, there were some DN glomeruli with a low MFSA but high spectral ratio, showing that IR spectral analysis can give information not available from the MFSA analysis alone. The 1030cm⁻¹ peak has been very well correlated with glycosylation. The increase of this peak would be expected in the case of diabetic nephropathy where the principal cause of tissue damage in diabetic patients is caused by uncontrolled level of glucose circulating in the blood. This leads to production of Advanced Glycation End products by non-enzymatic glycosylation typically by covalent bonding to proteins and lipids in tissues causing damage. The 1080cm⁻¹ peak is traditionally associated with DNA; however, we notice increases in the 1080cm⁻¹ peak in renal tissue structures that would not be expected to contain DNA. The 1080cm⁻¹ is also associated with glycogen and thus it is expected that we are again detecting tissue glycosylation (possibly of a different type).

Figure 3.

Histological and IR derived parameters are calculated for (A) mesangial fraction of surface area of glomeruli (MFSA) and (B) spectral absorbance ratio of 1030cm⁻¹ to 1080cm⁻¹ (1030:1080) from the mesangium for each glomerulus (n=68) from all thirteen patients. Each of these values per glomerulus are plotted as a separate bar and grouped separately per patient on the bar chart for the MFSA (Fig 3A) and the spectral absorbance 1030cm⁻¹:1080cm⁻¹ (Fig 3B). There are 5 or 6 glomeruli analyzed per patient. The mesangial IR spectra were obtained from an adjacent section of the same glomeruli analyzed histologically. The means were computed for diabetic nephropathy (DN), normal diabetic (NLD), and normal non-diabetic (NL) classes as shown by the dotted lines (Red=DN, Blue= NLD, Black=NL). (C) The 1030:1080 ratio was plotted (X-axis) against MFSA (Y-axis) for each glomeruli (each glomeruli is a single point). Each patient was assigned a unique symbol within each class (thus 5 or 6 symbols per patient). Quadrants were subjectively drawn to visualize separation. Patient 11 (inverted red triangle) which showed low mesangial expansion but had a high 1030:1080 spectral ratio was classified as having diabetic nephropathy based on EM which showed a mild thickening of the GBM and TBM.

Future work will focus on developing a model in-vitro system to better characterize what this biomarker is specifically related to in the tissue.

Finally, we applied a supervised multivariate data analysis technique called Linear Discriminant Analysis (LDA) to the IR data. LDA is a technique in which the entire spectral range is used for each glomerulus (3850–900cm⁻¹) and a known class assigned (Fig 4). LDA identifies sources of inter-group variance and maximizes discrimination between groups allowing for analysis of clustering based on biochemical similarity. Care was taken in selection of Principal Components to avoid potential overfitting in LDA (19). A very high level of separation between the three classes was achieved demonstrating that IR spectroscopy could identify spectral differences between the three groups (Fig 4). Interestingly, the NLDs (blue) were distinct from ND glomeruli (black) indicating that IR imaging is powerful enough to identify renal biochemical changes due to diabetes even in patients without histological changes.

Figure 4.

Linear Discriminant Analysis (LDA) was performed using the complete spectral data set for each of the features studied: (A) glomerular basement membrane (GBM), (B) tubular basement membrane (TBM) and (C) mesangium (M). Principal Component Analysis (PCA) was performed prior to performing LDA in order to reduce the dimensionality of the dataset while accounting for 99.99% of the variance in the data set. As seen in the LDA plots, the control (normal diabetic (NLD), normal non-diabetic (NL)) and diabetic nephropathy (DN) groups are very biochemically distinct. Therefore, separation between the groups are distinctly seen. And even within the control groups, it is possible to separate the NLD and NL groups. This shows that IR can very robustly distinguish not only between DN and control groups, but even between the two control groups used in this study

To determine whether the IR spectral signature could also be used in renal transplant, we studied 5 transplant patients with ESRD caused by diabetes. These patients with stable allograft function had undergone protocol had undergone protocol post-transplant biopsies, both early (3–6 moths post-transplant) and late (around 24 months post transplant) to check for subclinical complications. Three of five patients had diabetic nephropathy related histological changes in the late biopsy, with an example of one of the patients shown in Figure 5. All three had changes in the 1080:1030 spectral ratio (Fig 6A). The two biopsies without histologic changes of diabetic nephropathy showed no change in the spectral ratio.

Figure 5.

A patient with a history of diabetes mellitus type II complicated by diabetic nephropathy. Transplant biopsies were performed at 9, 21 and 36 months post-transplant. Light microscopy (top) and EM (bottom) shows changes associated with diabetic nephropathy are shown. Mesangium expansion is visible by PAS staining, while glomerular basement membrane thickening can be seen by EM. Arrows indicate glomerular basement membrane in the PAS image.

Figure 6.

(A) Spectral changes in mesangium from three patients associated with recurrent diabetic nephropathy (DN) and two patients without recurrent diabetic nephropathy (NLD). Increases in glycosylation (1030cm⁻¹) and DNA and glycosylation associated (1080cm⁻¹) peaks were observed in the patients with recurrent DN while patients without recurrent DN did not show any increases. This shows promise of tracking progression of diseased state. (B) Principal component analysis of histologically normal glomeruli from patients that either had recurrent diabetic nephropathy within three years (3 patients, red), and patients that did not have recurrence (two patients, blue). While all biopsies were histologically normal, there was clustering of biopsies from patients that later had recurrent DN. While the cohort for this study is small, this shows promise of an underlying biochemical signature that may be able to predict recurrence of DN prior to any histological DN changes. Please note the differential scaling of the X and Y axis of the plot. The age range of the patients was 28 to 59 years of age.

The early (3/6 month) biopsies had no clinical or histologic evidence of diabetic nephropathy. The average IR spectrum of the mesangium (measured in up to 6 glomeruli) from these biopsies did not show an altered spectral ratio. The entire spectrum (3850–900cm⁻¹) was analyzed using the unsupervised multivariate analysis technique, Principal Component Analysis (PCA). PCA converts every IR spectrum of a given sample into a single point and allows for analysis of clustering based on spectral variance where the closer the two points are the more the spectral similarity. Recurrent and non-recurrent diabetic nephropathy in late biopsies was associated with distinct clusters in the early biopsies (Fig 6B). This phenomenon suggests that underlying biochemical changes occur even in the absence of histologic evidence of diabetic nephropathy. Thus, IR imaging may allow for the detection of diabetic nephropathy in transplanted kidneys earlier than is morphologically evident.

DISCUSSION

IR imaging allows for the identification of a biochemical signature of diabetic nephropathy in renal biopsies from a single unstained tissue section. In addition, we can identify early biochemical changes associated with the recurrence of diabetic nephropathy in transplant patients prior to histological changes. Treatment modalities for diabetic nephropathy are not very effective, possibly because renal involvement is discovered too late. A new diagnostic tool that could detect very early change diabetic nephropathy could be useful for preventing further progression of disease, specifically in the post-transplant protocol serial biopsies setting. It could also be used as an intermediate endpoint in diabetic nephropathy recurrence prevention studies. Future studies are focused on using IR imaging for the assessment of interstitial fibrosis and exploiting the derived biochemical information to understand the mechanism of disease. In addition, the emergence of new laser based IR imaging systems has the potential for real-time imaging. We will also focus on the integration of spatial and morphological information from stained tissue sections with the spectral information from IR images to maximize the amount of diagnostic and prognostic information from tissue biopsies. It will also be important to expand the number of patients in these studies to allow for the training and validation of a predictive model with the requisite sensitivity and specificity (20).

METHODS

Subjects were characterized as diabetic based on a fasting glucose > 126mg/dL or any serum glucose level >200mg/dL. A clinical diagnosis of diabetic nephropathy was made when the estimated MDRD GFR was < 60mL/min/1.73m² and a spot urine albumin/creatinine ratio was >30mg/gm creatinine. Formalin fixed paraffin embedded (FFPE) tissue blocks were retrieved from the University of Illinois at Chicago, Tissue Bank (IRB approval protocol number 2014-0267). One section was placed on a standard glass slide and stained using Periodic acid Schiff (PAS) stain for histologic analysis (cut at 3μm) and an adjacent section was placed on an IR compatible substrate (cut at 4μm). Tissues for IR analysis were deparaffinized in hexane for 48 hours prior to imaging. The PAS stained slide was scanned with Aperio Scanscope CS (Nusslock, Germany) and digitally analyzed using Aperio Imagescope v11.2. The IR images were obtained with an Agilent FT-IR microscope (Santa Clara, CA) collected in transmission mode with an 36× collecting objective and 15× focusing objective with a pixel size of 2.2μm × 2.2μm similar to as previously described (14). The background was collected with 128 co-adds, and each image was collected with 64 co-adds. Spectral resolution of 8 cm⁻¹ was used over the range of 3850 to 900 cm⁻¹. The data was baselined using standard linear baseline correction, had noise reduction applied using the Minimum Noise Fraction approach and was normalized to the Amide I peak. Multivariate analysis was performed using Origin Pro 9.0 (Northampton, MA) for PCA and LDA. The native biopsy study consisted of 13 patients (age range 49 to 93), classed as NL (n=4), NLD (n=4) and DN (n=5) based on clinical, light microscopy and/or electron microscopy. Each patient has approximately 350 spectra extracted from each glomerulus, with either 5 or 6 glomeruli examined per patient. It took approximately 3 minutes to acquire an IR image of each glomerulus. The transplant biopsy study consisted of 5 patients with an early (approx. 3 months post-transplant) and late (approx. 24 months post-transplant) biopsy. Three of the patients had evidence of diabetic nephropathy recurrence in their late biopsies while two did not. The age range of the patients were 28 to 59 years of each. Each patient has approximately 350 spectra extracted from five glomeruli. The five spectra from each of the patients’ glomeruli were then averaged to form a single spectrum per patient of either early or late stage biopsy and were then subjected to further analysis (Fig 6).