Detection of Renal Hypoxia Configuration in Patients with Lupus Nephritis: A Primary Study Using Blood Oxygen Level‒Dependent MR Imaging

Renal microstructure and function are closely associated with homeostasis of oxygenation. Analyzing renal blood oxygen level‒dependent (BOLD) magnetic resonance imaging (MRI) examination results will provide information on the biological status of the kidneys. The current study was performed to explore the hypoxia mode of the entire renal parenchyma in patients with lupus nephritis (LN). Twenty-three adult patients with LN and eighteen healthy volunteers were recruited. R2* values were acquired through the use of the BOLD MRI analysis technique. The narrow rectangular region of interest was used to explore the hypoxia configuration in entire depths of renal parenchyma. Acquired sequential R2* data were fitted by using four categories of mathematic functions. The tendency of R2* data in both patients with LN and healthy volunteers was also compared through the use of repeated-measures analysis of variance. According to the detection results from the rectangular ROI of BOLD MRI, we found at least two pattern categories of R2* manifestation in the renal parenchyma. The first R2* pattern in the renal parenchyma displayed two different styles, shown in the cortex and the medulla. In the cortex or cortical‒medullary conjunction zone, the increment of R2* values concomitant with increasing depth renal parenchyma was slight, fluctuating in a small range. However, a conspicuous uptrend of R2* values was observed in the deeper medulla zone. Compared with the correlation between R2* values and renal depth of parenchyma, the slope of R2* values in the deeper medulla zone was steeper than that in the cortex zone. We call this pattern of R2* values the “sharp uptrend style.” The second R2* pattern in kidneys was different from the first pattern. The R2* values maintained a relatively stable level throughout the cortical and medullary zones. We call this pattern of R2* values the “flat uptrend style" (Figure 2). Based on these findings, we provided the definition of two styles of R2* values. If the highest R2* value exceeded 100% increasing magnitude of lowest R2* value, this sample was classified as sharp uptrend style. Correspondingly, if the highest R2* value did not exceed 25% increasing magnitude of lowest R2* value, this sample was classified as flat uptrend style. in the (Pillai’s statistic, P < 0.001; Wilks’ likelihood ratio, P 0.001; Hotelling-Lawley Trace criterion, P < 0.001; Roy’s Largest Root, P < 0.001). The statistical interaction effect between renal parenchyma layers and study groups was observed (Pillai’s Trace statistic, P = 0.004; Wilks’ likelihood ratio, P = 0.004; Hotelling-Lawley Trace criterion, P = 0.004; Roy’s Largest Root, P = 0.004).


Abstract
Background Renal microstructure and function are closely associated with homeostasis of oxygenation. Analyzing renal blood oxygen level-dependent (BOLD) magnetic resonance imaging (MRI) examination results will provide information on the biological status of the kidneys. The current study was performed to explore the hypoxia mode of the entire renal parenchyma in patients with lupus nephritis (LN).

Methods
Twenty-three adult patients with LN and eighteen healthy volunteers were recruited.
R2* values were acquired through the use of the BOLD MRI analysis technique. The narrow rectangular region of interest was used to explore the hypoxia configuration in entire depths of renal parenchyma. Acquired sequential R2* data were fitted by using four categories of mathematic functions. The tendency of R2* data in both patients with LN and healthy volunteers was also compared through the use of repeated-measures analysis of variance.
Results R2* data from the superficial cortex to deep medulla displayed two patterns, called a sharp uptrend style and a flat uptrend style. After sequential R2* data were fitted individually with the use of four mathematic formulas, the multiple-compartment Gaussian function showed the highest goodness of fit. Compared with two categories of R2* value styles, the R2* tendency of entire parenchyma in patients with LN was different from that in healthy volunteers.

Conclusions
Deep renal medullary oxygenation was not always overtly lower than oxygenation in superficial renal cortical zone. Renal parenchyma oxygenation manifestation could Unfortunately, there are no ideal methods of simultaneously detecting renal structure and pathophysiological status. Although renal biopsy specimen examination has been successfully applied in clinical practice for several decades, this technique is not considered to be ideal because of multiple inherent shortcomings, such as serious bleeding complications [4], incommodious field of view for inspection, and limited biopsy specimen locations. Therefore, a new, noninvasive, comprehensive assessment method is needed.
By using the paramagnetic properties of deoxyhemoglobin, the blood oxygen leveldependent (BOLD) magnetic resonance imaging (MRI) technique was deemed as a promising reliable and noninvasive inspection manner. Since the first renal BOLD MRI study was reported by Prasad in 1996 [5], investigators have increasingly explored renal oxygenation principles and tissue hypoxia mechanisms. During the past two decades, many renal physiological manifestations and formation mechanisms have been elucidated. These well-accepted renal physiological discoveries were corroborated with BOLD MRI results. For example, renal tissue in the medullary zone had a lower oxygen partial pressure and hypoxia gradient than in the entire kidney parenchyma [6]. The corresponding results were obtained from renal BOLD maps analyzed by use of the concentric objects (CO) [7] and twelvelayer concentric objects (TLCO) techniques [8]. Although substantial valuable discoveries have been reported by nephrologists and radiologists worldwide, several issues have prevented BOLD MRI from extensive clinical use and have frustrated initial research enthusiasm. The first issue concerns the origin of the renal BOLD signals. It is well known that massive physicochemical or biological factors are involved in renal BOLD signal formation. These factors include regional blood oxygen supplementation, tissue oxygen consumption, hydration status, hematocrit, and salt and medicine intake [9][10][11]. Therefore, analysis of the BOLD signal value alone cannot determine which solitary or multiple factors are the key factors. A second problem relates to the method of renal BOLD signal analysis. Although the acquisition of renal BOLD MR images is easily performed worldwide, there is no consensus as to how to analyze the BOLD MR images. Different analysis methods could lead to significantly different results. For example, Milani et al. investigated the difference in the R2* value between patients and control subjects and did not find any difference in R2* with the use of the conventional regional ROI technique.
However, significant differences in R2* values in cortical layers were revealed when the TLCO technique was adopted in the same patient cohort [8]. Overcoming these native deficiencies of BOLD MRI may assist in further understanding renal tissue oxygenation.
The main purposes of this study were to understand renal parenchyma hypoxia configuration and to develop a new BOLD image analysis method. We also compared tissue hypoxia characteristics between healthy people and patients with LN.

MRI techniques
We acquired renal BOLD-MRI images by using a 3.0-T scanner (GE Discovery™ 750 3.0T; General Electric; USA). The detailed imaging parameters can be referred to our previous study [16] Image analysis Renal tissue oxygenation distribution was displayed by visualized R2* map by processing with FUNCTOOL program. Renal coronal anatomical plane with largest area was selected as analyzed section. The region of interest (ROI) was set as a 1 pixel (width) × 50 pixels (height) rectangle. One tip of the ROI was located on the renal cortical surface. Another tip of the ROI looked out on the renal hilus ( Figure   1 The effects of intergroup (LN vs healthy control) and intragroup (different depths of renal parenchyma) were compared. The interaction effect was also compared.

Curve-fitting analysis
To describe the R2* configuration throughout the cortex and medulla, the curvefitting analysis was selected to explore the detected R2* data. Four curve fit types (polynomial, power, exponential, and Gaussian) were chosen for data exploration.
The goodness of fit was assessed with the use of multiple parameters including R 2 , sum square error (ESS), and root-mean-square error (RMSE). All analyses were carried out by using the Curve Fitting Toolbox of MATLAB R2014a (MathWorks Inc.).

Results
General clinical and pathological results of patients with lupus nephritis A total of twenty-three patients with lupus nephritis were recruited in our current Manifestation of R2* values in renal parenchyma from superficial cortex to deep medulla According to the detection results from the rectangular ROI of BOLD MRI, we found at least two pattern categories of R2* manifestation in the renal parenchyma. The first R2* pattern in the renal parenchyma displayed two different styles, shown in the cortex and the medulla. In the cortex or cortical-medullary conjunction zone, the increment of R2* values concomitant with increasing depth renal parenchyma was slight, fluctuating in a small range. However, a conspicuous uptrend of R2* values was observed in the deeper medulla zone. Compared with the correlation between R2* values and renal depth of parenchyma, the slope of R2* values in the deeper medulla zone was steeper than that in the cortex zone. We call this pattern of R2* values the "sharp uptrend style." The second R2* pattern in kidneys was different from the first pattern. The R2* values maintained a relatively stable level throughout the cortical and medullary zones. We call this pattern of R2* values the "flat uptrend style" (Figure 2). Based on these findings, we provided the definition of two styles of R2* values. If the highest R2* value exceeded 100% increasing magnitude of lowest R2* value, this sample was classified as sharp uptrend style.
Correspondingly, if the highest R2* value did not exceed 25% increasing magnitude of lowest R2* value, this sample was classified as flat uptrend style.

Analysis of two patterns of R2* values shows a difference between patients with LN and healthy volunteers
In order to analyze two styles of R2* values in renal parenchyma, the extreme R2* data including highest R2* average values with maximal range and lowest R2* average values with minimal range were selected from each study subject. We selected 8 clusters highest extreme R2* data and 8 clusters lowest extreme R2* data in each research subject. Because the height of the rectangular ROI was 50 pixels, we divided the renal parenchyma into 50 layers. The average R2* values in each renal parenchyma layers are shown in Figure 3.
There was a very small difference in the R2 value curves between the LN group and the control group when the depth of renal parenchyma was less than 25 layers.
However, the slope of the R2* curve in the LN group was lower than that in the control group at a depth of 2550 renal layers. We compared the sharp uptrend style pattern of R2* values in both the LN group and the control group by using RM-ANOVA. Because Mauchly's test showed that covariance structure was not satisfied with sphericity assumption (P < 0.001), multivariate ANOVA was chosen for further data. The statistical difference in R2* values between renal parenchyma layers was observed in both the LN group and the healthy volunteer group (Pillai's Trace statistic, P < 0.001; Wilks' likelihood ratio, P < 0.001; Hotelling-Lawley Trace criterion, P < 0.001; Roy's Largest Root, P < 0.001). The statistical interaction effect between renal parenchyma layers and study groups was also observed (Pillai's Trace statistic, P = 0.004; Wilks' likelihood ratio, P = 0.004; Hotelling-Lawley Trace criterion, P = 0.004; Roy's Largest Root, P = 0.004). Similar results were found for the flat uptrend style pattern of R2* data. In the superficial zone of renal parenchyma (less than 25 layers), the location of the R2* curve in the LN group was lower than that in the control group. However, the difference in R2* values in the deep medullary zone was not distinct. RM-ANOVA results showed a statistic difference between the two R2* curves. Multivariate

Discussion
One of the primary aims was to understand the relationship between tissue oxygenation and renal parenchyma depth. Our current study revealed two categories of renal oxygenation manifestation patterns. One pattern, the sharp uptrend style, hasbetter tissue oxygenation with a slight fluctuation in the superficial layers of the kidney. Subsequently, a sharp increment of deoxygenation was observed in the deep layers of the renal parenchyma. Although this pattern has similar characteristics with the well-known renal tissue oxygenation feature, such as higher deoxygenation in the medulla than in the cortex [19], a distinct two-phase oxygenation feature was different from that in many previous studies [20][21][22]. This sharp uptrend style of renal R2* values might correlate with a sophisticated relationship between tissue partial pressure of oxygen (PaO 2 ) and oxyhemoglobin saturation dissociation curve. Tissue oxyhemoglobin saturation could be maintained in a stabilized level when PaO 2 is above 60 mmHg, whereas there is a steep gradient with a sharp decrease in oxyhemoglobin when the PaO 2 is less than 26.6 mmHg. Previous research has shown that PaO 2 in the majority of cortexes is usually higher than 60 mmHg, whereas medullary PaO 2 rarely exceeds 26.6 mmHg [23].
Another relevant reason was interrelated with 2,3-diphosphoglycerate (DPG)mediated oxyhemoglobin affinity. A lessening in oxyhemoglobin affinity has been deemed as an important physiological adaptive response to conditions in which oxygen delivery is impaired. The increased oxyhemoglobin affinity may actually impair tissue oxygenation [24]. Moreover, another pattern of R2* values, the flat uptrend style, was also found in our current study. We found that the R2* values fluctuated in a narrow range throughout the depth of renal parenchyma. R2* values in the deep medulla were only slightly higher than those in the superficial cortex.
We even found no discrepancy in R2* level between the superficial renal zone and the deep renal zone, in sporadic samples. This phenomenon implies that tissue oxygenation in the deep medullary zone is not always lower than that in the superficial cortical zone. This discovery is a challenge to the recognized opinion that had been testified by many studies.
Unfortunately, we could not detect renal parenchyma histological and cellular physiological conditions simultaneously. The potential explanation of these two patterns was only based on reasonable ratiocination. In renal cortical zone, the oxygen tension is more variable because of the fluctuation of renal blood perfusion.
The existence of an arterial-to-venous oxygen shunt mechanism can maintain the average partial pressure of oxygen (pO2) at approximately 30 mmHg. Schurek et al and Welch et al verified that oxygen tension in renal vein exceeded that in glomerular capillaries and that in the efferent arteriole, respectively [25,26]. These findings confirmed the existence of oxygen shunt, which was the renal oxygenation homeostatic mechanism. The arterial-to-venous oxygen shunt mechanism maybe one of reasons why deoxyhemoglobin can be detected in the well blood perfusion cortical zone. However, the oxygen supplement in medullary zone was obvious lower than that in cortical zone, not to mention high local oxygen consumption for tubular reabsorption. Other compensative mechanisms such as prostaglandins, nitric oxide (NO), and adenosine were involved. These compensative mechanisms continuously adjust medullary tubular transport activity to the limited available oxygen supply, acting by both enhancement of regional blood flow and downregulation of distal tubular transport, particularly in medullary thick ascending limbs (mTAL) [23]. We speculated the gradient of deoxyhemoglobin in medulla perhaps preserved at similar level in cortex. On the other hand, another pattern of R2* value called "sharp uptrend style" could be also found. It was implied that the quantity of deoxyhemoglobin increased significantly. This R2* values manifestation was observed not only in patients with lupus nephritis but also in healthy volunteers. We speculated several possible pathophysiological mechanisms were involved. Firstly, glomerular hyperfiltration is one condition that increases oxygen demand in renal parenchyma. When glomerular filtration rate is exorbitance, more plasma is also filter out. This condition results in sodium overload in tubular lumen and requires more energy for the reabsorption of sodium by tubular cells [27]. Secondly, oxidative stress can result tissue hypoxia by superoxide radicals decreasing the bioavailability of nitric oxide. Reduced tissue nitric oxide leads to vasoconstriction and decrease in regional blood flow. Furthermore, increased oxidative stress can evaluate oxygen consumption in kidney, possibly via the effects on tubular transporters [28,29]. Thirdly, loss of peritubular capillaries or reduction in peritubular capillary flow may give raise to renal hypoxia. Previous studies on animal models have showed that tubulointerstitial injury is associated with the distortion and loss of peritubular capillaries [30]. Activation of renin-angiotensin system (RAS) and imbalance of vasoactive substances lead to vasoconstriction and lessen peritubular capillary flow [27].
Our current findings were different from both earlier BOLD MRI studies and direct measurement of tissue oxygen levels with oxygen electrodes animal studies. The possible explanation might come from the principle rationale of BOLD MRI.
Oxygenated hemoglobin is diamagnetic and has no magnetic moment, whereas deoxygenated hemoglobin is paramagnetic and can affect the apparent spin-spin relaxation time (T2*) of adjacent water protons. The reciprocal of T2*, which is designated the apparent or transverse relaxation rate, is also called R2* which reflected the amounts of deoxygenated hemoglobin in tissue [31]. Therefore, when we assess renal tissue oxygenation by R2* parameter, we actually detect deoxygenated hemoglobin rather than partial oxygen pressure. Because of special relationship between hemoglobin oxygen saturation and oxygen partial pressure, which can be expressed as Hill's equation, the BOLD MRI technique is suitable for low oxygen tension environments in which the O2 hemoglobin dissociation curve has maximal effect. In other words, deoxyhemoglobin based R2* parameter will not reflect microelectrode-based oxygen partial pressure index synchronously.
Moreover, local redistribution of renal blood flow (RBF), oxygen consumption, or peritubular capillary network changes can also affect corticomedullary R2* discrepancy. Several previous studies have also shown that furosemide, water loading, or many renal diseases can reduce or even eliminate the normal corticomedullary differences in R2* [32,33]. Another reason why previous study did not find this exception may derive from the ROI mode, which could remarkably affect the detective R2* value. For example, the TLCO technique was recognized as the preferable manner by which renal R2* values were measured in the past few years [8]. The entire renal parenchyma of the R2 map image was divided into 12 consecutive equivalent depths of the zone. Subsequently, the R2* value of each pixel was detected, and the average R2* value in each zone was also calculated.
Because of lower variability and higher stability in repeated measures, this renal R2* acquired technique was well accepted by many investigators. However, the calculated average R2* value was usually prone to affect by higher medullary R2* data instead of those lower R2* data. Under this circumstance, the phenomenon of lower R2* level in the deep medulla was concealed by this technique. There was no plausible explanation for this new discovery as observed in our study. We thought that the microstructure and physiological status in that deep medulla zone might discriminate those renal medullary tissues where the typical oxygenation pattern was easily observed.
Another result of our study involved exploring the possibility of describing renal parenchyma R2* patterns by mathematic models. Some investigators thought that the renal tissue oxygenation status was influenced by multiple factors, such as local blood supplementation and tissue oxygen consumption. Therefore, detected R2* values in each pixel from anywhere in the BOLD images were always the integrated results of multiple physiological factors. The key problem was that the precise participation proportion of each factor was not well understood. However, we believed that information on all involved physiological factors with similar to the encrypted message was still sealed in these R2* maps. We still hope to unlock this sealed information with the right decoding method. By testing the multiple mathematic functions, we found that Gaussian function had the best capability to fit practical R2* data. Moreover, both the sharp uptrend style data and the flat uptrend style data fit the Gaussian function well with two or more compartments. Although there was no reliable evidence to verify the corresponding relationship between renal biological factors and compartments of fit Gaussian functions, we hypothesized that those multiple encrypted and perplexing physiological messages in the R2* data could be transformed into another mode. Perhaps we will decode the encrypted biological messages that were composed in R2* data by studying fit Gaussian functions.
Our study also compared renal histological data with R2* values. In proliferative LN such as type III or type IV, lower renal tissue R2* values usually correspond with severe tubular damage. The inverse manifestation was observed in non-proliferative LN in which mild tubular injuries could be confirmed with renal biopsy samples [34].
Despite statistical discrepancy of R2* values detected between the two groups in both sharp uptrend and flat uptrend style data, the substantial R2* difference was very small. We speculated that at least two factors were involved and that the pathogenesis was similar to that in patients with renal artery stenosis. Hansell et al.
found that kidneys usually had increased renal blood flow instead of regional pO2 when the renal parenchyma was under severe hypoxia conditions. The increased renal blood flow also led to GFR increments that subsequently induced more filtered sodium in the renal tubules. The reinforced tubular sodium reabsorption was the main reason for increased oxygen consumption [35]. Due to R2* values indicated magnitude of deoxyhemoglobin according to the principle of BOLD imaging, lower R2* in cortex implied the better cortical oxygenation. Morphology assessment of renal biopsy samples showed tubulointerstitial inflammation or atrophy could be been in patients with proliferative lupus nephritis. We speculated these pathological injuries might weaken tubular transport activity, which subsequently reduced oxygen consumption and decreased the R2* values. Our surmise was also confirmed by our previous study. Furthermore, faded GFR also led to downward R2* due to lessened solute delivery for tubular transport. These possible mechanisms listed above might explain the tendency of R2* in Figure 3 that both crossover of two R2* curves in medulla and separation of two R2* curves in cortex could be detected simultaneously. Our research team also considered that the integrity of renal physiological function in healthy volunteers was superior to that in patients with lupus nephritis. The preserved glomerular and tubulointerstitial construction and function were prone to oxygen consumption, and integrate peritubular capillary would transport more hemoglobin into renal tissues. We speculated that this was one of the reasons why higher R2* values were found at deep medullary zone in sharp upturn style and superficial cortical zone in flat upturn style.
We also developed a novel ROI method called the "narrow rectangle" or "virtual probe" to measure renal R2* values. This was the principal reason why we devised and applied this special ROI technique. Renal microstructure displays distinct directional characteristics and spatial heterogeneity. Renal parenchyma oxygenation features are based on its histological anatomy foundation. The wellknown opinion that the degree of renal blood oxygen saturation in the cortical zone is always higher than that in the medullary zone is based on previous ROI analysis methods such as regional ROI selection [ handled nonconsecutive crude R2* data by regional ROI selection or acquired calculated mean R2* results by CO and TLCO methods. These inherent shortcomings remained an obstacle to the procurement of the explicit oxygenation tendency throughout the renal parenchyma. Contrary to previous conventional ROI analysis methods, our new analysis technique revealed that deep medullary oxygenation was not always overtly lower than cortical oxygenation. This unusual discovery challenged the well-known opinion.
Our current study still had several shortcomings. First, our study merely investigated renal R2* signal instead of other functional MR signals such as ADC, synchronously. We only knew the existing oxygenation condition, whereas we did not understand the precise mechanism of hypoxia status formation. Second, the meaning of fitted Gaussian functions had not yet been explained or proved by other physiological studies, and we only hypothesized that fitted Gaussian formulas represent renal hypoxia outline in the kidney cortical and medullary zones. Third, the narrow rectangular ROI analysis method that was adopted in our study had not been tested by many researchers. Because the highly observer-dependent property of previous conventional ROI analysis methods has been proved, we did not know the new ROI analysis technique had the highly reproducible and lower variant characteristic.

Conclusions
Deep renal medullary oxygenation was not always overtly lower than that in the superficial renal cortical zone. Renal parenchyma oxygenation manifestation could Declarations Ethics approval and consent to participate General Hospital prior to the commencement of the study. Written informed consent for participation was obtained from the patient or a legal representative.

Consent for publication Not applicable
Availability of data and material The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.