Efficacy of Urine Asymmetric Dimethylarginine Concentration to Predict Azotemia in Hyperthyroid Cats After Radio‐Iodine Treatment

ABSTRACT

Background

Hyperthyroidism can mask concurrent chronic kidney disease in cats, and no accurate biomarkers are available to predict which cats will develop renal azotemia after radioiodine (¹³¹I) treatment.

Hypothesis/Objectives

To evaluate the potential of serum and urinary metabolites and metabolite ratios to predict post‐¹³¹I renal azotemia in hyperthyroid cats.

Animals

Hyperthyroid cats (n = 31), before and (3–12 months) after treatment with ¹³¹I at the Faculty of Veterinary Medicine (Ghent University, Belgium).

Methods

Retrospective study. Optimized and validated feline extraction and analysis protocols were employed for metabolic profiling of urine and serum samples using ultra‐high performance liquid chromatography–high‐resolution mass spectrometry. A dual strategy of cross‐validated univariate and penalized multivariate logistic regression was applied to determine predictivity (i.e., area under the curve [AUC], accuracy, sensitivity, and specificity) of individual biomarkers and panels.

Results

All hyperthyroid cats were non‐azotemic before ¹³¹I administration. After ¹³¹I treatment, 7 cats became persistently (≥ 2 timepoints) azotemic while 24 remained non‐azotemic. Urinary asymmetric dimethylarginine (ADMA) was identified as a pivotal predictor of post‐¹³¹I azotemia in both univariate and multivariate modeling. When employed as a standalone biomarker, an AUC of 0.851, accuracy of 0.903, sensitivity of 0.714, and specificity of 0.958 were achieved. While pre‐treatment USG was significantly different (P = 0.002) between both groups, it did not show enhanced prediction over ADMA, nor in multivariate modeling.

Conclusions and Clinical Importance

Urinary ADMA can accurately predict post‐¹³¹I azotemia in hyperthyroid cats becoming euthyroid after ¹³¹I treatment. These findings can aid clinicians in managing owner expectations and modify treatment plans.

Abbreviations

¹³¹I

radioactive iodine

ADMA

asymmetric dimethylarginine

AUC

area under the curve

CKD

chronic kidney disease

Cr

creatinine

DDAH

dimethylarginine dimethylaminohydrolase

eNOS

endothelial nitric oxide synthase

FDR

false discovery rate

HRMS

high‐resolution mass spectrometry

LASSO

least absolute shrinkage and selection operator

m/z value

mass‐to‐charge ratio

NO

nitric oxide

PCA‐X

principal component analysis

Q‐orbitrap

quadrupole orbitrap

ROC

receiver operating characteristic

R _t

retention time

SDMA

symmetric dimethylarginine

T4

total thyroxine

TSH

thyroid stimulating hormone

UHPLC

ultra‐high performance liquid chromatography

UPC

urine protein–creatinine ratio

UPW

ultrapure water

USG

urinary specific gravity

1. Introduction

Hyperthyroidism is a prevalent endocrine disorder in elderly cats [1], characterized by the excessive production of thyroid hormones. This can lead to various systemic effects, including increased glomerular filtration rate (GFR) and renal blood flow [2, 3]. In combination with a loss of muscle mass [4], there is a decline in serum creatinine (Cr) concentrations, thereby masking potential concurrent kidney function impairment. After treatment of hyperthyroidism, the previously undetected chronic kidney disease (CKD) is revealed [5, 6].

Radioiodine (¹³¹I) has been successfully used in the treatment of hyperthyroidism [7, 8, 9, 10, 11, 12]. However, predicting which cats will develop post‐¹³¹I azotemia remains challenging, yet relevant for tailoring treatment plans and potentially improving clinical outcomes. There is a shorter survival time in hyperthyroid cats developing hypothyroidism and azotemia after ¹³¹I treatment; this is not the case for euthyroid azotemic cats [13]. Nevertheless, uncovering renal deterioration before therapy remains of importance, as it allows owners to be informed and take timely measures. Recent advances in metabolomics, which involve the comprehensive analysis of metabolites in biological samples, offer a promising avenue for identifying biomarkers linked to disease states and treatment responses [14]. Therefore, metabolomic profiling can provide insight into the metabolic alterations that precede the onset of post‐¹³¹I azotemia.

This study utilizes optimized and validated feline urine and serum extraction and analysis protocols for targeted metabolomics [15], to identify predictive biomarkers of post‐treatment azotemia in hyperthyroid cats treated with ¹³¹I. By comparing the metabolic profiles of cats that develop post‐treatment azotemia with those that maintain normal renal function, we seek to uncover metabolic alterations that predict the development of post‐¹³¹I azotemia.

2. Materials and Methods

2.1. Study Sample and Inclusion Criteria

Study cats (n = 31) were retrospectively selected from a cohort of hyperthyroid cats (n = 75), treated with ¹³¹I at the Small Animal Department from the Faculty of Veterinary Medicine at Ghent University between December 2020 and January 2023. Initially, subjects were recruited for a different study [16]. Cats were included in this study if they met the inclusion criteria as described below and if residual samples were available. Owners were informed about the study objectives and procedures, and provided written consent. Cats were examined before ¹³¹I treatment (T0), and control visits were performed 3 (T3), 6 (T6), and 12 (T12) months after ¹³¹I treatment. In order to assess general health, information obtained from the anamnesis was evaluated in conjunction with a physical examination, including blood pressure using the Doppler ultrasonic method, bloodwork, and urinalysis [17, 18, 19, 20]. More specifically, in‐house analyses were performed for whole blood count (IDEXX ProCyte Dx [IDEXX Laboratories Inc., Westbrook, ME, USA]), routine biochemistry (IDEXX Catalyst Dx Chemistry Analyzer [IDEXX Laboratories Inc., Westbrook, ME, USA]), dipstick testing (IDEXX VetLab UA‐analyzer [IDEXX Laboratories Inc., Westbrook, ME, USA]), measurement of urinary specific gravity (USG; Master SUR/NM handheld refractometer [Atago, Tokyo, Japan]), and microscopic sediment analysis. Measurement of the urine protein–creatinine ratio (UPC; Roche Cobas c702i [Roche Diagnostics, Diegem, Belgium]) and urine bacterial culture (only performed at T0) were performed in a commercial laboratory. After completion of the follow‐up period, samples were sent in batch for measurement of total thyroxine (T4) and thyroid stimulating hormone (TSH), using a validated feline electrochemiluminescence‐immunoassay (Immulite 2000 XPi [Siemens, Erlangen, Germany]). At T0, a thyroid scintigraphy using technetium 99m pertechnetate (^99mTcO₄ ⁻) was performed in all cats before ¹³¹I treatment to determine functional status, location, and size of the thyroid [21]. Hyperthyroidism was diagnosed based on increased T4 concentrations (T4 > 3.5 μg/dL), TSH below the limit of detection (< 0.03 ng/mL) and/or scintigraphic confirmation of hyperthyroidism through an increased thyroid‐to‐salivary ratio (T/S) [22]. Cats were treated intravenously with individualized doses of ¹³¹I, based on T4, clinical disease severity, qualitative assessment, and quantitative T/S assessment of the 99mTc‐pertechnetate thyroid scintigraphy [23].

Cats were included in the study if the cat underwent a minimum of two control visits following ¹³¹I treatment, in order to confirm the classification of cats in the azotemic or non‐azotemic group. Only samples from post‐¹³¹I euthyroid cats (T4 between 1.1 and 3.5 μg/dL, TSH ≤ 0.3 ng/mL) were retrospectively included in this analysis, of which 7 originated from post‐¹³¹I azotemic cats and 24 from cats remaining non‐azotemic. For the azotemic group, urine and serum samples were collected at T0 and the first timepoint at which renal azotemia was detectable (Figure 1). For the non‐azotemic group, samples were selected from T0 and T3. Development of post‐¹³¹I azotemia was diagnosed based on a compatible history and physical examination combined with appropriate laboratory findings. More specifically, azotemia was defined using a cut‐off of 2.0 mg/dL for Cr, combined with a USG < 1.035, similar to previous studies describing post‐treatment azotemia in hyperthyroid cats [13, 16, 24, 25, 26, 27, 28]. Moreover, as hyperthyroidism influences renal function [3, 29], pretreatment evaluation of azotemia based on the IRIS guidelines would not be appropriate in the presence of comorbidities such as hyperthyroidism. Medication was not an exclusion criterium, except for anti‐thyroid medication, which had to be discontinued at least 1 week before ¹³¹I treatment. Moreover, cats with pre‐existing azotemia at T0 or concomitant systemic diseases (except for post‐¹³¹I CKD) were excluded from this study sample.

FIGURE 1.

Flowchart illustrating the number of included cats in each group. All cats were euthyroid after ¹³¹I treatment. For the cats developing post‐¹³¹I azotemia, the timepoint at which renal azotemia was detected for the first time is presented as well. All cats underwent at least two control visits after ¹³¹I treatment to confirm their azotemic or non‐azotemic status.

2.2. Sample Collection

Urine was collected through ultrasound‐guided cystocentesis using a 5 mL syringe with a 22‐G needle. Blood samples were obtained after a 30‐min topical application of EMLA cream (AstraZeneca, Södertälje, Sweden) via jugular venipuncture using a 10 mL syringe with a 23‐G needle [30]. Centrifugation of serum was performed for 5 min at 2°C (2190g). Supernatants were aliquoted into plastic Eppendorf tubes and stored at −80°C until metabolomics analysis.

2.3. Reagents and Chemicals

Deuterium‐labeled internal standards (Data S1) and analytical reference standards (Data S1) were purchased from Sigma‐Aldrich (St‐Louis, MO, USA), LGC standards (Teddington, London, UK), and MedChemExpress MCE (Princeton, NJ, USA). The Arium 611UV purification system from Sartorius (Göttingen, Germany) was used to obtain ultrapure water (UPW). Solvents including acetone, acetonitrile (ACN), and methanol (MeOH) were obtained from Fisher Scientific (Loughborough, UK) and VWR International (Darmstadt, Germany).

2.4. Extraction and Analytical Methods

Based on the results of our previous metabolomics study on early biomarkers for CKD prediction in healthy cats, a panel of urinary (metabolites: n = 64, metabolite ratios: n = 36) and serum (metabolites: n = 79, metabolite ratios: n = 45) analytical standards was composed for targeted analysis (Data S1). Extraction and analysis were performed according to the optimized and validated protocols for feline urine and serum samples of Vanden Broecke et al. [15] (Supporting Information). An ultra high‐performance liquid chromatography coupled with quadrupole orbitrap high‐resolution mass spectrometry (UHPLC‐Q‐Orbitrap HRMS) method was employed. A Vanquish Horizon UHPLC system (Thermo Fisher Scientific, San José, CA, USA) was applied for chromatographic separation, with an Acquity HSS C18 column T3 (1.8 μm, 150 mm × 2.1 mm; Waters, Manchester, UK) kept at 45°C. A gradient elution program at a flow rate of 0.4 mL/min was employed, using a binary solvent system with UPW and ACN, both acidified with 0.1% formic acid. A Thermo Fisher Scientific Exploris 120 Q‐Orbitrap benchtop mass spectrometer (San José, CA, USA) was employed, with a heated electrospray ionization source (HESI II) in polarity switching mode.

2.5. Data Processing

For accurate annotation, the mass‐to‐charge ratio (m/z value), retention time (R _t), and ¹³C/¹²C isotope ratio of targeted metabolites were compared to analytical standards. The maximum R _t and mass shift were set at 2.5% and 5 ppm, respectively, and the minimum signal‐to‐noise ratio at 10 [31, 32, 33]. Targeted processing was performed using Xcalibur 3.0 (Thermo Fisher Scientific, San José, CA, USA). Data preprocessing involved data normalization using the average intensity of two consecutive quality control (QC) samples after every 10 extracts to correct for instrumental drift during analysis. Log transformation and pareto scaling were applied for multivariate analysis to induce normality and standardize the range of peak intensities [34].

2.6. Statistical Analysis

Statistical analysis was performed using R (version 2023.03.0) and SIMCA 18 software (Umetrics, Malmö, Sweden) [35]. Pairwise comparisons were performed as a first exploratory step of the dataset, using a Welch two‐sample t‐test for parametric data or Wilcoxon rank‐sum test for non‐parametric data. Adjusted p‐values (< 0.05), which were corrected for the false discovery rate (FDR), were calculated using the Benjamini–Hochberg procedure [36]. Principal component analysis (PCA‐X) was performed to assess natural clustering among samples and identify potential outliers. Both univariate and multivariate least absolute shrinkage and selection operator (LASSO) logistic regression modeling were performed. This dual approach allowed us to compare the contribution of individual biomarkers and their combined effect as a panel in predicting post‐¹³¹I azotemia. Univariate and LASSO regression were performed using the R caret package, via the train() function with the method set at glm and glmnet, respectively. A repeated five‐fold cross‐validation approach using the trainControl() function was applied to evaluate performance (i.e., area under the curve [AUC], accuracy, sensitivity, and specificity), and to tune the penalization parameter lambda (λ) for LASSO regression [37, 38]. In order to calculate accuracy, sensitivity, and specificity, two approaches were employed: (i) by means of a default cut‐off value of 0.500 for classification based on predicted probabilities, and (ii) by determining the optimal cut‐off point using the Youden Index, which maximizes the sum of sensitivity and specificity (Youden Index = sensitivity + specificity − 1) [39]. Moreover, the following formulas were applied for the aforementioned metrics: accuracy = (true positives + true negatives)/total, sensitivity = true positives/(true positives + false negatives), specificity = true negatives/(true negatives + false positives). Additionally, based on previous studies [24, 40], we also assessed whether a USG < 1.035 could predict post‐¹³¹I azotemia and improve performance when added to the feature selection phase during LASSO regression.

3. Results

3.1. Study Sample and Demographics

A total of 31 cats were retrospectively selected from the initial cohort of 75 hyperthyroid cats treated with ¹³¹I. Urine and serum samples were analyzed from post‐¹³¹I euthyroid cats, of which 7 were azotemic and 24 non‐azotemic after treatment (Figure 1). Characteristics of both study groups are presented in Table 1 and Data S1. Although all hyperthyroid cats were non‐azotemic before ¹³¹I treatment, there was a significant difference in Cr (p‐value = 0.003) and USG (p‐value = 0.002) between both groups.

TABLE 1.

Demographics of all cats in the azotemic and non‐azotemic groups. Sample size, breed, and sex (including neuter status) are reported as n; age, creatinine (Cr), urinary specific gravity (USG), total thyroxine (T4), and thyroid stimulating hormone (TSH) values are reported as median (range). Statistical significance was assessed using the Wilcoxon rank‐sum test for all parameters except gender, which was evaluated by the Chi‐squared test.

	Azotemic group	Non‐azotemic group	p
N	7	24
Breed	DSH = 5 Ragdoll = 1 Siamese = 1	DSH = 22 Norwegian Forest Cat = 1 Siamese = 1
Sex (M, MC, F, FS)	1, 4, 1, 1	3, 8, 1, 12	0.446
Age at inclusion (years)	13.3 (11.2–16.7)	12.7 (7.9–18.5)	0.493
USG
Before 131I	1.020 (1.014–1.038)	1.039 (1.027–1.054)	0.002
After 131I	1.016 (1.012–1.022)	1.040 (1.015–1.050)	< 0.001
Cr (mg/dL)
Before 131I	1.1 (0.8–1.9)	0.8 (0.4–1.4)	0.003
After 131I	2.3 (2.1–3.7)	1.3 (1.0–1.9)	< 0.001
T4 (μg/dL)
Before 131I	9.4 (4.4–15.0)	11.0 (4.7–15.0)	0.414
After 131I	1.5 (1.1–3.0)	1.9 (1.1–3.0)	0.112
TSH (ng/mL)
Before 131I	0.03 (0.03–0.03)	0.03 (0.03–0.05)	0.176
After 131I	0.06 (0.03–0.29)	0.06 (0.03–0.21)	0.772

Note: Post‐¹³¹I renal azotemia was defined using a cut‐off of 2.0 mg/dL for Cr, combined with a USG < 1.035 on at least two post‐¹³¹I control visits while being euthyroid (T4 between 1.1 and 3.5 μg/dL, TSH ≤ 0.3 ng/mL).

Abbreviations: DSH = domestic shorthair; F = female intact; FS = female spayed; M = male intact; MC = male castrated; N = number of animals in group.

3.2. Metabolic Alterations Before and After Treatment of Hyperthyroidism With ¹³¹I

All metabolites and metabolite ratios from the a priori selected targeted panel (Data S1) could be detected in both study groups. Before ¹³¹I treatment, 10 urine metabolites exhibited significant (FDR corrected p‐value < 0.05) alterations between the two groups, while for serum, there were no significant differences (Table 2, Figure 2). After ¹³¹I treatment, 23 metabolites and metabolite ratios were significantly altered in urine, while only one metabolite (i.e., 2‐hydroxyethanesulfonate) demonstrated significance in serum (Table 2, Figure 2). No significant alterations were observed either before or after treatment when ratios of metabolites in both matrices (i.e., serum‐to‐urine ratio and urine‐to‐serum ratio) were calculated.

TABLE 2.

Significant (FDR corrected p‐value < 0.05) up‐ and downregulated metabolites and metabolite ratios in urine (U) and serum (S) of azotemic cats compared to non‐azotemic cats, before and after treatment of hyperthyroidism with ¹³¹I. Results are sorted according to their p‐value.

Matrix	Metabolite	p	FC
Before 131I
U	Creatinine	0.027	0.573
U	ADMA	0.027	0.520
U	Guanine	0.038	0.361
U	Cytidine	0.048	0.663
U	Lysine	0.048	0.631
U	Ornithine	0.048	0.485
U	5‐Hydroxyindole‐3‐acetic acid	0.048	0.637
U	Xanthurenic acid	0.048	0.515
U	Adenine	0.049	0.459
U	Tryptophan	0.049	0.501
After 131I
U	Creatinine	0.003	0.557
U	Leucine	0.003	0.547
U	ADMA	0.003	0.559
U	Isoleucine	0.003	0.358
U	Xanthurenic acid	0.005	0.534
U	Betaine	0.010	0.543
U	3‐Hydroxykynurenine	0.010	0.521
S	2‐Hydroethanesulfonate	0.010	1.842
U	Arginine/ornithine	0.011	2.250
U	N‐acetylneuraminate	0.018	1.776
U	5‐Hydroxyindole‐3‐acetic acid	0.018	0.590
U	Creatine	0.019	0.491
U	Fumaric acid	0.020	0.308
U	Xanthosine	0.020	0.553
U	Putrescine	0.033	0.503
U	Pantothenic acid	0.036	0.284
U	Phenylalanine	0.038	0.522
U	Serotonine	0.038	0.251
U	Adenosine	0.039	0.529
U	Gluconic acid	0.039	1.555
U	Kynurenine	0.039	0.616
U	N‐acetylleucine	0.039	0.354
U	2‐Hydroethanesulfonate	0.039	1.855
U	Creatine/guanidinosuccinate	0.039	0.167

Abbreviations: ADMA = asymmetric dimethylarginine; FC = fold change.

FIGURE 2.

Alterations in serum and urinary metabolome of hyperthyroid cats remaining non‐azotemic (n = 24) and becoming azotemic (n = 7) after ¹³¹I treatment. PCA‐X plots show few outliers, and an overlap of the study groups can be noted before as well as after ¹³¹I treatment in both serum (A and B, respectively) and urine (E and F, respectively). Volcano plots show significantly altered metabolites and metabolite ratios before (n = 10) and after (n = 23) ¹³¹I treatment in urine (G and H, respectively), with no altered metabolites in serum before treatment (C) and only one metabolite after ¹³¹I treatment (D).

3.3. Metabolic Alterations That Predict Post‐ ¹³¹I Azotemia

Both univariate and multivariate LASSO regression were employed to evaluate the predictive power of individual biomarkers and biomarker panels. Univariate logistic regression provided 11 significant (p‐value < 0.05) metabolites and two metabolite ratios in urine (Table 3). For serum, only two metabolites and one metabolite ratio showed significance, while for the ratios of both matrices, only leucine demonstrated a significant p‐value (Table 3). Cross‐validated performance metrics indicated urinary asymmetric dimethylarginine (ADMA) as the metabolite with the highest predictive capacity, demonstrating a high AUC (0.851), accuracy (0.903), specificity (0.958), and moderate sensitivity (0.714) for both thresholds (Table 4, Figure 3). Consequently, the probability of a cat with hyperthyroidism developing post‐¹³¹I azotemia can be calculated based on the obtained logistic regression function, as demonstrated below:

$$\begin{array}{c}p\left(\text{post}{‐}^{131}\mathrm{I}\text{azotemia}\right)=\frac{1}{1+\exp \left(-z\right)}\hfill \end{array}$$

with z = 4.191–4.248 × ADMA, and exp = the exponential function based on Euler's number. To illustrate this, consider a cat that developed azotemia following ¹³¹I treatment in this study with a pre‐¹³¹I ADMA value of 0.552, resulting in a p of 0.864 ($p=\frac{1}{1+{\mathrm{e}}^{-\left(4.191-4.248\times 0.552\right)}}$). Comparing this value to the defined threshold (0.500 for the default calculation or 0.505 for the Youden Index), this cat is likely to develop post‐¹³¹I azotemia as p exceeds the threshold.

TABLE 3.

Metrics of univariate logistic regression for the prediction of post‐¹³¹I azotemia using metabolites and metabolites ratios in urine (U), serum (S), and ratios of matrices, sorted according to their p‐value. Statistical significance was evaluated using the Wald test. Only metabolites and metabolites ratios with a p‐value < 0.05 are presented.

Matrix	Metabolite	β 0	Estimate	SE	OR	p
U	ADMA	4.191	−4.248	1.565	69.999	0.007
U	Arginine/ADMA	−5.060	3.743	1.569	0.024	0.017
U	Creatinine	3.617	−4.914	2.092	136.149	0.019
U	Arginine/ornithine	−3.811	1.528	0.668	0.217	0.022
U	Cytidine	3.790	−5.094	2.249	162.991	0.024
U	Ornithine	1.312	−2.740	1.219	15.488	0.025
U	Leucine	2.137	−2.306	1.043	10.033	0.027
U/S	Leucine	−3.112	1.164	0.541	0.312	0.031
U	3‐Hydroxykynurenine	3.271	−3.730	1.746	41.688	0.033
U	Adenine	3.351	−4.809	2.263	122.604	0.034
U	Adenosine	0.661	−1.720	0.828	5.587	0.038
S	Betaine/dimethylglycine	3.321	−6.300	3.044	544.725	0.038
S	Betaine	1.951	−4.135	2.045	62.479	0.043
U	Xanthosine	2.258	−2.797	1.387	16.391	0.044
U	Lysine	1.773	−2.602	1.297	13.497	0.045
U	Xanthurenic acid	1.350	−1.111	0.557	3.038	0.046
S	2‐Hydroethanesulfonate	−4.321	2.491	1.256	0.083	0.047

Abbreviations: β ₀ = intercept; ADMA = asymmetric dimethylarginine; OR = odds ratio; SE = standard error; U/S = urine‐to‐serum ratio.

TABLE 4.

Performance metrics (i.e., AUC, accuracy, sensitivity, and specificity) of individual metabolites and metabolite ratios in urine (U), serum (S), and ratios of matrices. Both a default threshold of 0.500 and an optimized threshold based on the Youden Index were evaluated to predict post‐¹³¹I azotemia. All significant (p‐value < 0.05, Wald test) metabolites and metabolite ratios are presented, sorted according to the highest AUC.

Matrix	Metabolite	AUC (CI)	Threshold	Acc (CI)	Sens	Spec
U	ADMA	0.851 (0.774–0.928)	0.500	0.903 (0.845–0.945)	0.714	0.958
			0.505	0.903 (0.845–0.945)	0.714	0.958
U	Ornithine	0.827 (0.714–0.939)	0.500	0.858 (0.793–0.909)	0.486	0.967
			0.279	0.852 (0.786–0.904)	0.857	0.850
U	Creatinine	0.821 (0.738–0.905)	0.500	0.819 (0.750–0.876)	0.514	0.908
			0.241	0.800 (0.728–0.860)	0.829	0.792
U	Adenine	0.819 (0.749–0.889)	0.500	0.806 (0.735–0.865)	0.429	0.917
			0.103	0.652 (0.571–0.726)	0.971	0.558
U	Cytidine	0.800 (0.719–0.882)	0.500	0.781 (0.707–0.843)	0.371	0.900
			0.200	0.748 (0.672–0.815)	0.857	0.717
S	Betaine/dimethylglycine	0.766 (0.684–0.849)	0.500	0.766 (0.688–0.832)	0.400	0.882
			0.098	0.586 (0.502–0.667)	1.000	0.455
U	Xanthurenic acid	0.762 (0.676–0.848)	0.500	0.742 (0.666–0.832)	0.171	0.908
			0.374	0.819 (0.750–0.876)	0.714	0.850
U	Xanthosine	0.762 (0.673–0.851)	0.500	0.806 (0.735–0.865)	0.343	0.942
			0.290	0.755 (0.679–0.820)	0.686	0.775
U	Adenosine	0.759 (0.666–0.851)	0.500	0.768 (0.693–0.832)	0.257	0.917
			0.344	0.819 (0.750–0.876)	0.657	0.867
U	Leucine	0.750 (0.657–0.842)	0.500	0.813 (0.742–0.871)	0.429	0.925
			0.203	0.684 (0.604–0.756)	0.714	0.675
U	Lysine	0.744 (0.667–0.821)	0.500	0.735 (0.659–0.803)	0.171	0.900
			0.166	0.671 (0.591–0.744)	0.857	0.617
U	3‐Hydroxykynurenine	0.741 (0.652–0.830)	0.500	0.768 (0.693–0.832)	0.314	0.900
			0.159	0.632 (0.551–0.708)	0.800	0.583
U	Arginine/ADMA	0.725 (0.612–0.838)	0.500	0.813 (0.742–0.871)	0.457	0.917
			0.405	0.839 (0.771–0.893)	0.571	0.917
U	Arginine/ornithine	0.718 (0.593–0.843)	0.500	0.903 (0.845–0.945)	0.571	1.000
			0.501	0.903 (0.845–0.945)	0.571	1.000
S	Betaine	0.713 (0.627–0.800)	0.500	0.731 (0.651–0.801)	0.229	0.891
			0.220	0.669 (0.586–0.745)	0.857	0.609
U/S	Leucine	0.686 (0.559–0.813)	0.500	0.766 (0.688–0.832)	0.286	0.918
			0.236	0.800 (0.726–0.862)	0.629	0.855
S	2‐Hydroethanesulfonate	0.656 (0.546–0.766)	0.500	0.786 (0.710–0.850)	0.286	0.945
			0.442	0.800 (0.726–0.862)	0.429	0.918

Abbreviations: Acc = accuracy; ADMA = asymmetric dimethylarginine; AUC = area under the curve; CI = 95% confidence interval; Sens = sensitivity; Spec = specificity; U/S = urine‐to‐serum ratio.

FIGURE 3.

Longitudinal evolution and performance of urinary asymmetric dimethylarginine (ADMA) and the biomarker panel. (A) Boxplot representation of changing ADMA concentrations in hyperthyroid cats remaining non‐azotemic (n = 24) and becoming azotemic (n = 7) after ¹³¹I treatment. Statistical significance was evaluated using the Welch two‐sample t‐test. (B) Analysis of variable importance of included metabolites and metabolite ratios in urinary biomarker panel (n = 4). Each metabolite is assigned a weight between 0 and 100, with higher values indicating greater influence within the model. (C and D) Receiver operating characteristic (ROC) curve of ADMA and urinary panel for classification of post‐¹³¹I azotemia. The AUC with 95% CI is presented as well.

LASSO regression was used to select metabolites for inclusion in a predictive biomarker panel. The tuning parameter lambda (λ) was trained by repeated k‐fold cross‐validation (k = 5) based on accuracy, resulting in a more stringent (i.e., greater λ‐values) penalization for both matrices separately than when combined (Table 5).

TABLE 5.

Performance metrics (i.e., AUC, accuracy, sensitivity, and specificity) of LASSO models, involving metabolites and metabolite ratios in urine (U) and serum (S) and both matrices combined. Both a default threshold of 0.500 and an optimized threshold based on the Youden Index were evaluated to predict post‐¹³¹I azotemia.

Matrix	N	λ	AUC (CI)	Threshold	Acc (CI)	Sens	Spec
U	4	0.126	0.865 (0.788–0.942)	0.500	0.827 (0.757–0.882)	0.343	0.967
				0.259	0.877 (0.815–0.825)	0.743	0.917
S	7	0.132	0.620 (0.517–0.723)	0.500	0.768 (0.688–0.832)	0.086	0.982
				0.256	0.669 (0.586–0.745)	0.486	0.727
Combined	6	0.091	0.852 (0.777–0.927)	0.500	0.841 (0.764–0.891)	0.400	0.973
				0.158	0.772 (0.695–0.838)	0.886	0.736

Abbreviations: λ = regularization parameter lambda; Acc = accuracy; AUC = area under the curve; CI = 95% confidence interval; N = metabolites in panel; Sens = sensitivity; Spec = specificity.

For prediction of post‐¹³¹I azotemia using a biomarker panel, the highest performance was obtained for the urine matrix (Figure 3, Table 5). A panel of four urinary metabolites was constructed, all of which exhibited decreased concentrations in the post‐¹³¹I azotemic group. More specifically, in the logistic regression function to estimate the probability of developing post‐¹³¹I azotemia using urine metabolites, z was now defined as −1.489 − 0.853 × ADMA − 0.294 × adenine − 0.581 × ornithine − 0.069 × 5‐hydroxyindole‐3‐acetic acid.

Additionally, USG (either ≥ or < 1.035) provided a moderate predictive capacity in univariate logistic regression (i.e., OR = 12.0, p‐value = 0.03, AUC = 0.661, accuracy = 0.710, sensitivity = 0.857, specificity = 0.667). When adding USG to the feature selection phase in multivariate LASSO regression, this variable was eliminated from the predictive model through the penalization process, thereby not enhancing the performance of the initial models.

4. Discussion

This study demonstrates the potential of metabolomic profiling of predictive biomarkers of post‐¹³¹I azotemia in hyperthyroid cats. More specifically, both univariate and multivariate LASSO regression identified urinary ADMA as a key metabolite for the prediction of post‐¹³¹I azotemia. When used as a single biomarker, an AUC of 0.851, accuracy of 0.903, sensitivity of 0.714, and specificity of 0.958 were obtained. A multi‐biomarker panel involving urinary metabolites (n = 4) demonstrated a similar performance.

Numerous studies on the prediction of azotemia in hyperthyroid cats treated with ¹³¹I have been performed over the past years. It has been suggested that symmetric dimethylarginine (SDMA) could serve as a predictor of post‐¹³¹I azotemia, although sensitivity is poor (15.4%–33.3%) [40, 41]. Moreover, mildly elevated serum SDMA in hyperthyroid cats can normalize after the resolution of hyperthyroidism, thus necessitating careful interpretation [41, 42]. Cats with a suboptimal USG before ¹³¹I administration are more likely to develop post‐treatment azotemia [24]. Indeed, our study also showed a significant (p‐value = 0.002) difference in USG between hyperthyroid cats remaining non‐azotemic after ¹³¹I treatment and those developing azotemia. Furthermore, a moderate performance (AUC = 0.661, accuracy = 0.710) was obtained through univariate logistic regression, resulting in a sensitivity of 0.857 and specificity of 0.667, similar as in a study on the influence of ¹³¹I therapy, masked azotemia, and iatrogenic hypothyroidism on the urine concentrating ability in cats with hyperthyroidism (0.861 and 0.652, respectively) [24]. Nonetheless, as cats with an USG > 1.035 can also develop post‐¹³¹I azotemia, it is not feasible to determine a cut‐off value for clinical practice. In addition, studies have been conducted on GFR [5, 43], Cr [43, 44], T4 [43, 44], UPC [26, 43], blood urea nitrogen (BUN) [43, 44], cystatin C (CysC) [27, 45, 46], retinol binding protein (RBP) [43, 47], vascular endothelial growth factor (VEGF) [28], N‐acetyl‐β‐d‐glucosaminidase (NAG) [48, 49], and fibroblast growth factor 23 (FGF23) [50]. However, these studies revealed negative predictive outcomes, shortcomings, practical limitations, or were insufficiently validated, rendering these biomarkers unsuitable for predictive application in clinical practice.

Our study suggests urinary ADMA as a potential predictor for post‐¹³¹I azotemia. ADMA is an endogenously formed amino acid, generated with its enantiomer SDMA by monomethylation (i.e., protein arginine methyltransferases [PRMTs]), and hydrolysis of protein‐bound l‐arginine. Although SDMA has been extensively studied in CKD in cats [51, 52, 53, 54, 55, 56, 57, 58], including post‐¹³¹I renal azotemia [40, 41, 42, 59], ADMA has only been described in the context of CKD in cats in the work of Jepson et al. [60]. More specifically, they demonstrated a moderate correlation (r = 0.608, p < 0.001) between plasma ADMA and Cr concentrations in cats (n = 69) with CKD [60]. Human and rodent studies have shown that both SDMA and ADMA are uremic toxins [61, 62], playing a role in the pathogenesis of several diseases, including CKD [63, 64, 65, 66]. Moreover, ADMA has been identified as a biomarker of increased mortality and rapid progression of human CKD [64, 67, 68]. In contrast to SDMA, which correlates well with GFR, ADMA is predominantly metabolized into dimethylamine and l‐citrulline by dimethylarginine dimethylaminohydrolase (DDAH) enzymes, with only 10%–20% of ADMA being excreted through the kidneys [69, 70, 71]. Jepson et al. [60] suggested that an enzyme equivalent to DDAH might exist in cats as well for ADMA metabolization. It has been demonstrated that a reduction in DDAH metabolism, resulting from oxidative stress and reduction of tubular mass, represents the primary factor contributing to elevated plasma ADMA levels in human and rodent CKD, independent of changes in GFR [66, 72, 73]. In addition, ADMA is an endogenous competitive inhibitor of endothelial nitric oxide synthase (eNOS), causing reduced nitric oxide (NO) formation [64, 74]. NO is an important vasodilator which exerts a profound impact on numerous renal and cardiovascular functions, including the modulation of renal autoregulation, tubular fluid and electrolyte transport, vascular tone, and blood pressure [75]. Besides competitive inhibition, ADMA also decreases eNOS through inhibition of its phosphorylation [76].

Unlike circulating ADMA, few studies have assessed the potential of urinary ADMA in predicting the onset or progression of (post‐¹³¹I) human CKD [77]. Moreover, the mechanism of ADMA excretion in the presence of renal damage has not yet been elucidated. Although renal excretion is not the main route of elimination, as illustrated in several studies on human, rat, and rabbit metabolism of ADMA [78, 79, 80, 81], significantly decreased urinary ADMA concentrations were measured in the azotemic group before (p‐value = 0.027, FC = 0.520) and after (p‐value = 0.003, FC = 0.559) ¹³¹I treatment. The extent to which renal excretion and therefore reduced renal clearance is responsible for the decreased urinary ADMA concentrations in azotemic cats remains unclear. Nevertheless, a similar trend was noted in a human study on the effect of renal function on ADMA homeostasis [77], in which GFR and urinary ADMA measurements were performed on 130 healthy kidney donors, before and after donation, as a model of isolated renal function impairment. A significant (p‐value < 0.001) decrease was revealed in both parameters at a median of 1.64 (1.61–1.87) months after donation.

ADMA has been associated with various clinical conditions in human medicine, emphasizing the need for additional studies to assess its specificity [63, 70]. Furthermore, a reference interval for ADMA in hyperthyroid cats should be established to predict post‐¹³¹I azotemia, as well as a validated commercial assay for ADMA measurement. In particular, commercial ELISA kits are already available for serum, plasma, and urine of humans and mice [82, 83]. However, such immunoassays might not be feasible for larger multi‐biomarker panels. While liquid chromatography‐mass spectrometry (LC‐MS) analyses are powerful, they are time‐consuming and labor‐intensive in terms of sample preparation, which impedes their scalability for clinical use. However, such advanced diagnostics broadens the scope of clinical health screenings, offering the potential to detect early renal deterioration alongside a spectrum of other conditions. Despite these obstacles, MS platforms have been used as the gold standard for several assays in feline medicine, including serum 25‐hydroxyvitamin D and SDMA [84, 85]. Moreover, emerging ambient ionization techniques present a transformative opportunity by eliminating extensive sample preparation and enabling direct analysis of crude samples [86, 87, 88].

One limitation of this study was the limited number of cats. Of all cats that came from the initial cohort of hyperthyroid cats (n = 75), 21 developed post‐¹³¹I azotemia. However, only post‐¹³¹I euthyroid cats with renal azotemia on at least two control visits after ¹³¹I were selected for this study (n = 7). This decision was based on two arguments, that is, to eliminate the influence of a decreased thyroid function (i.e., iatrogenic overt or subclinical hypothyroidism) on potential biomarkers, and to avoid inclusion of iatrogenic hypothyroid azotemic cats, as they do not provide an accurate representation of true azotemic CKD [3, 89]. Additionally, ultrasonography to detect renal abnormalities and measurement of GFR were not performed. Since group classification was based on the presence or absence of renal azotemia, it cannot be excluded that cats with early renal deterioration, however non‐azotemic, were included in the non‐azotemic control group. However, as Cr is not an early biomarker for renal deterioration [90], samples from cats that remained non‐azotemic following ¹³¹I treatment were selected from the first control visit after ¹³¹I administration, being T3. This approach ensured us that their non‐azotemic status could also be verified at later visits, namely T6 and T12. In the case of post‐¹³¹I azotemic cats, the first time point at which azotemia was diagnosed was selected. Moreover, data from control visits provided additional confirmation of the absence of other potential, previously masked systemic diseases. Finally, one cat from the non‐azotemic group was treated with benazepril before and after ¹³¹I treatment, while two cats (one from each group) received gabapentin several hours before sample collection, only during the control visits after ¹³¹I treatment. It is noteworthy that Sibal et al. [91] reported that ACE inhibitors can reduce plasma ADMA concentrations, which could have influenced ADMA concentrations in that specific cat in our study.

Despite extensive efforts to identify accurate biomarkers for the prediction of azotemia in hyperthyroid cats treated with ¹³¹I, progress in this area has remained rather limited. In our study, utilizing optimized and rigorously validated metabolomics protocols, we identified urinary ADMA as a reliable predictor of post‐¹³¹I azotemia. These findings could aid clinicians in managing owner expectations and modifying treatment plans.

Disclosure

Authors declare no off‐label use of antimicrobials.

Ethics Statement

Ethical approval was obtained from the ethical committee of the Faculties of Veterinary Medicine and Bioscience Engineering (EC2017/64) at Ghent University. Authors declare human ethics approval was not needed.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Data S1. Supporting Information.