A new analysis approach for single nephron GFR in intravital microscopy of mice

Background: Intravital microscopy is an emerging technique in life science with applications in kidney research. Longitudinal observation of (patho-)physiological processes in living mice is possible in the smallest functional unit of the kidney, a single nephron (sn). In particular, effects on glomerular filtration rate (GFR) - a key parameter of renal function - can be assessed. Methods: After intravenous injection of a freely filtered, non-resorbable, fluorescent dye in C57BL/6 mice, a time series was captured by multiphoton microsopy. Filtration was observed from the glomerular capillaries to the proximal tubule (PT) and the tubular signal intensity shift was analyzed to calculate the snGFR. Results: Previously described methods for snGFR analysis relied on two manually defined measurement points in the PT and the tubular volume was merely estimated in 2D images. We present an extended image processing workflow by adding continuous measurement of intensity along the PT in every frame of the time series using ImageJ. Automatic modelling of actual PT volume in a 3D dataset replaced 2D volume estimation. Subsequent data analysis in R, with a calculation of intensity shifts in every frame and normalization against tubular volume, allowed exact assessment of snGFR by linear regression. Repeated analysis of image data obtained in healthy mice showed a striking increase of reproducibility by reduction of user interaction. Conclusions: These improvements in image processing and data analysis maximize the reliability of a sophisticated intravital microscopy technique for the precise assessment of snGFR, a highly relevant predictor of kidney function.


Introduction
Glomerular filtration rate (GFR) is a key parameter of kidney function and deviations from normal GFR are a hallmark of renal diseases 1,2 . GFR describes the filtration of substances from blood in the glomerular capillaries, to the primary urine in the tubular system of the kidney. Therefore, changes in GFR serve to monitor disease progression 1,2 . GFR is also measured in animal models to study effects of pharmacological intervention on kidney function 3 . Advances in intravital imaging and multiphoton microscopy allow repetitive assessment of GFR and morphological changes in the smallest functional unit of the kidney -the nephron 3-5 . Longitudinal imaging of single nephrons (sn) enable direct correlation of structural and functional data 3-5 .
After intravenous injection of the freely filtered, non-resorbable, fluorescent dye LuciferYellow (LY), a time series was captured by multiphoton microsopy. Filtration was observed from the glomerular capillaries to the proximal tubule (PT) and the tubular signal intensity shift is analyzed to calculate the filtration rate. Translated to an image processing task, this can be generalized as the flow rate in a tube. Previous methods for this analysis 3,4 relied on two manually annotated measurement points in the PT and stereotypic estimation of PT volume in 2D images. Since results we obtained with this approach were highly variable, we expanded the analysis of image data via 3D modelling with open source software, to increase overall reproducibility and reliability of the analysis when comparing renal function of different experimental groups.

Animal experiments
Animal experiments were performed in accordance with the Federation of European Laboratory Animal Science Associations (FELASA) Guidelines for the Care and Use of Laboratory Animals and the Federal Law on the Use of Experimental Animals in Germany and approved by the ethical review committee at the Landesdirektion Sachsen (license DD-24. 1-5131/338/37). For microscopy, male, 10-12 week old C57BL/6 mice were prepared as previously described 5,6 . In brief, a titanium abdominal imaging window (AIW) covered with a coverslip is surgically implanted above the kidney. The kidney is glued to the coverslip with cyanoacrylate glue before securing the AIW by tightening the skin in the AIW groove. Microscopy was performed one day after AIW implantation.
A custom-built temporary intravenous catheter (polyethylene tubing #587360 by Science Products GmbH with 0.3×12mm needle) was placed in the lateral tail vein. Fluorescent dyes were administered into the tail vein prior (Hoechst, AngioSpark) or during (LuciferYellow) microscopy (detailed information in Table 1).
All efforts were made to ameliorate harm to animals. Imaging (including injections of the fluorescent dyes) and the implantation is done under isoflurane anaesthesia. The image data of the five animals presented for the comparison of the extended workflow with the previous workflow in this manuscript were generated previously as part of an independent experiment (license DD-24.1-5131/338/37).

Microscopy
Imaging was performed on an upright Leica SP8 multiphoton laser scanning microscope at the Core Facility Cellular Imaging. Settings for signal acquisition are summarized in Table 2.
Image and data analysis Image processing and analysis was done in ImageJ 7-9 (1.53c) with 3D ImageJ Suite 10 and Bio-Formats 11 for the use of 3D image processing plugins and the Bio-Formats Importer. Data analysis was performed in R 12 (4.0.2), with RStudio 13 (1.2.5033) including ggplot2 14 (including dependencies) installed as additional library. The script executed the ImageJ macro from command line and subsequently analyzed and visualized the results. A detailed description of the algorithm is associated with the scripts on GitHub 15 .
The line region of interest (ROI) set for the extended workflow to manually define direction and position of the proximal tubule (PT) was also used to determine the two measuring points (beginning and end of line) for analysis of image material based on the previously described approach 3,4 . Tubular diameter was calculated as the mean of five manually measured diameters.

Results
In the time series acquired after application of LY, a line ROI was set to manually define the position and direction of the measurement. Along this ROI, x-y plots measured the dye intensity in the PT in every frame ( Figure 1) and numerical results were saved.
For the automatic 3D modelling of PT volume the z-stack of the same field of view was acquired. Additional channels (Ch3: AngioSpark -vessels, Ch4: Hoechst -nuclei, Figure 2A) were subtracted plane by plane from Ch2 (target channel,  For visualization of the resulting data, the signal intensity along the ROI is plotted for these sample frames on the right. As the LY moves through the PT, the measured signal intensity shifts as well.  , the cumulative volume was measured, providing a conversion of position to volume. B) Numerical data underlying the x-y plots was saved and used to subsequently plot changes of signal intensity over time for every position (and converted to cumulative volume) along the line ROI. The dashed line represents the threshold value at which the corresponding volume of the proximal tubule (PT) was approximated for every frame. C) Using linear regression the snGFR could be calculated as the volume with the intensity threshold at the frames of interest (after conversion from µm³ per frame to nl per minute). Regression line is displayed with 95% confidence interval. The colour codes for the position along the PT (blue -beginning, red -end).
LY intensity) to remove spectral bleed-through artifacts ( Figure 2B). With the 3D watershed, the PT was segmented ( Figure 2C, 3D-model) and saved for visual verification. The cumulative PT volume was measured over the distance along the ROI and plotted in subsequent data analysis ( Figure 3A). The position is now recalculated to the cumulative PT volume at each point along the ROI. From intensity measurements a threshold intensity was set to the turning point of fluorescence intensity over time at every volume (maximum slope, Figure 3B). The volume with this intensity was approximated in each frame and used for linear regression ( Figure 3C, intersect of horizontal threshold at every frame with intensity curves). The slope of the regression line equals the snGFR after conversion of µm³ per frame to nl per minute. Together with information about PT length, PT volume and R-squared the results were summarized and saved in a data table.
Repeated analysis (five times) of 15 individual glomeruli by the same researcher showed that results obtained with the presented workflow had higher consistency (lower intrasample variance, CV=10.35%) compared to the previous approach (CV=38.75%, Figure 4). Due to the high variance with the previous approach a direct correlation of the workflows was not possible; however, the final result -the mean snGFR -was comparable (previous workflow: 1.71±0.91, extended workflow: 1.70±0.78) and a two-sample Kolmogorov-Smirnof test of both result vectors showed that the distributions were not statistically different (p=0.4662). Numerical results of the repeated analysis with both workflows are listed in Table 3.

Conclusions
The progressive development of microscopy techniques like measurement of snGFR in experimental animals needs to be accompanied by improvements in analysis algorithms to use their full potential. In this manuscript we present a workflow by extending an existing analysis method via 3D modelling, for increased reproducibility, accuracy, but also transparency in the measurement of snGFR. By reducing user interaction, intrasample variance was markedly improved.   Additionally, the automatically saved user input and intermediate results (z-stack of watershed of PT as shown in Figure 2C and graphs in Figure 4) for every analyzed dataset provide full possibility to verify every analysis step. These results can be used to objectively evaluate the measurement. Although the snGFR in this manuscript was very low for healthy animals compared to previously published values 3 , the range was comparable in both methods and not an artifact produced by the workflow but more likely caused by the general experimental setup.
Taken together, this workflow extension contributes to an overall improvement of the interpretation of snGFR measurements. Applied to experimental data this can cumulate in a higher power to detect statistically significant differences between experimental groups and even decrease the necessary sample size, thus having an impact on animal welfare. Open Peer Review nephron (sn) and glomerular filtration rate (GFR) in mice. In the results section the authors describe an extended image processing workflow with ImageJ/Fiji and "R". Conclusively, this study shows an improved manner for image processing in order to study and analyse snGFR. The study seems technically sound to me.

Data availability
However, Figure 1 is to my opinion not optimal. If separated from the main manuscript text, it is difficult to understand the presented results in this figure. Therefore, I would advise the authors to revise the figure description.

If applicable, is the statistical analysis and its interpretation appropriate? I cannot comment. A qualified statistician is required.
Are all the source data underlying the results available to ensure full reproducibility? Yes

Are the conclusions drawn adequately supported by the results? Yes
Competing Interests: No competing interests were disclosed.
Reviewer Expertise: Multiphoton microscopy and Light-sheet microscopy with application to mouse modes.

I confirm that I have read this submission and believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard, however I have significant reservations, as outlined above.
Author Response 08 Jun 2021 Friederike Kessel, University Hospital Carl Gustav Carus at the Technische Universität Dresden, Fetscherstraße 74, Dresden, Germany We thank Dr. Schüth for their reviewer report and the helpful remarks on Figure 1 and the insufficient figure captions. We agree that without the context of the entire article, the caption should contain additional information.
We address this issue in a new version of the manuscript and hope Figure 1 is now more understandable.
The article is based on a false claim and hypothesis that the current published methods for SNGFR measurement is unreliable and produce high variability. Therefore, its scientific validity is questionable. There seems to be one major and one minor issue with this work. The major issue concerns the unclear overall significance of this technical advance because it provides only a minor incremental advance in the field. The claim that the conventional workflow does not provide reproducible and reliable results is not valid. There are many advantages of the conventional manual analysis method and when performed correctly by the published methods and meticulous observer the results are highly reproducible and have low variability. It is puzzling how the application of the conventional and new approach can produce over 5-fold differences in results from the same sample (Table 3). In my opinion, there are more stressing issues with the measurement of SNGFR, that would warrant the development of such advanced analytical tools. For example, SNGFR is not constant over time as compared to global GFR due to vasomotor (myogenic) tone and tubuloglomerular feedback (TGF). This results in significant alterations in GFR on the single nephron level by every 5-10 seconds (myogenic) or 30-50 seconds (TGF). Therefore, this is a more important issue in SNGFR variability than intra-observer variability in the analysis of a single timepoint SNGFR measurement.
The minor issue is the quality of the preparations shown in the supplement material as previously raised by Dr. Molitoris. There seems to be a significant amount of blebs and cell debris flowing in many tubular segments. All numerical SNGFR values shown in table 3 are below the physiological range. These suggest that the animals were not in physiological healthy conditions. Using these datasets for the current analytic development work is not optimal.

Is the work clearly and accurately presented and does it cite the current literature? Yes
Is the study design appropriate and is the work technically sound? Partly

If applicable, is the statistical analysis and its interpretation appropriate? Partly
Are all the source data underlying the results available to ensure full reproducibility? Yes

Are the conclusions drawn adequately supported by the results? No
ROI to determine the position and direction for the measurement of the intensity change across the time course. Then the entire 3D volume is segmented in a separate z-stack. The analysis is smart and tries to use all available information in their experimental setup to ensure a robust analysis. The generation of the 3D volume based on a 3D watershed uses the information from different channels to ensure a robust segmentation. Using a regression analysis makes sure that not only 2 points alone will contribute to the final measurement of the snGFR. I think this is a good image and data analysis approach to reduce measurement variability and increase statistical power.
The workflow runs as R script that also calls a Fiji macro. The interaction runs via sequential GUI prompts. This increases the ease of use.
The article is, in general, well written and contains most of the important information to understand the method and how it compares to previously used methods. The detailed description of the algorithm is sometimes a bit confusing but one can understand the rational of the authors. My main problem is with the lack of documentation that allows one to access and implement the tools. Here are my points for revision:

MAJOR:
The algorithm makes sense. The description in the text and figure legends is however a bit hard to understand.
This sentence is particularly unclear: "The position with this intensity was approximated in each frame and used for linear regression ( Figure 3C)". I guess what the authors wanted to express is that the volume was approximated at this position. Then the approximated volume was plotted over time and based on that linear regression was performed?  Figure 3B&C corresponds I guess to the positions? This is not explained anywhere. In Figure 3A first µm³ is used and then in Figure 3C nl?
MAJOR: I downloaded the material and it took me about 4 tries to get the scripts to work correctly.
Here are the key impediments: Does not execute on ubuntu 20.04: the R script uses functions that only work under Windows. This limitation needs to be explicitly stated. This also abolishes the advantage of cross platform tools such as Fiji and R.

1.
The 3D watershed is not explicitly stated as dependency of Fiji. It needs to be clearly stated what needs to be installed and how. 2.
The direction of the ROI is important but this was not clear from the documentation. 3.

MAJOR:
The Documentation word file provided is not helpful for actually using the scripts. It rather contains a code documentation that has some directions of using the program included. People with little expertise have no clear guidance for the usage and the important settings of the usage are entirely lost in all the detail.
The usage needs to be documented separately from the code. 1.
The actual interaction with the program needs to be documented also via screenshots. 2.
The important settings need to be explained clearly and in sufficient detail. 3.

MAJOR:
That one needs to draw the ROI in the direction of the wave was not really obviously documented or it got lost in the complexity of the code documentation. Please use screenshots or describe with words.

MAJOR:
How the data needs to be acquired and structured for this workflow to function is not explained anywhere. The prompt for selecting a corresponding z-stack made initially zero sense. Since it was not clear that the .lif file must contain the multichannel time series AND the z-stack. Are the channels settings hard coded then? If the analysis is inflexible in its data input (which can be ok), it needs to be mention explicitly as an important prerequisite.

MAJOR:
Reproducing the workflow using the provided .lif file resulted mostly in snGFR that were in a similar range. But still off. Maybe drawing the ROIs seems to be still an important source of variability. It would be good if there would be an easy way to load and visualize the ROIs provided by the authors. This shows easily how the authors intend users to set ROIs. One can load them via the ROI Manager during the GUI interactions but this produces an error later on: Composite selections cannot be converted to lines. in line 520:

(called from line 193) run ( "Area to Line" <)>
Maybe it would also be good to document in words along example screenshots how one best should set the ROI.
MAJOR: I am not in the kidney field. Maybe certain statements are common knowledge there and it is practice not to cite them. But the following statements in the introduction would strike me as requiring citations: "Therefore, changes in GFR serve to monitor disease progression." ○ "GFR is also measured in animal models to study effects of pharmacological intervention on kidney function." First of all we thank the reviewer for the extensive report and constructive feedback. We agree that a lack of documentation, guidance and also support information notably impaired the usability of the workflow. In this context, the patience of the reviewer to implement the algorithm is highly appreciated.
Since most of the remarks were directly linked with lack of documentation, we uploaded detailed instructions to the GitHub repository and updated the associated release on Zenodo. This documentation now includes: A paragraph on the structure of the raw data ○ System requirements ( Since this analysis was only recently developed and experiences when applying it to different image data are still limited we are determined to continuously expand the documentation and troubleshooting suggestions. We recognize that there is also room for improvement for the programming itself, regarding the limitation to Windows and hardcoded requirements of the raw data. We plan to support the gradual expansion of the workflow to be more adaptable -and applicable -in the future (as mentioned in the documentation).
Since the reviewer pointed out that some of the descriptions in the manuscript and figure legends were hard to understand, we rephrased some points. We hope it is now more understandable. Description of Figure 3C  We also included the table with the numerical results as shown in Figure 4. Finally, the statements on GFR and methods in intravital microscopy in the introduction can be supported with references that were already used in other contexts in the manuscript. Therefore, we additionally refer to them in the introduction.

Competing Interests:
No competing interests were disclosed.