INFLUENCE OF IRON CHELATION ON R1 AND R2 CALIBRATION CURVES IN GERBIL LIVER AND HEART

Abstract

MRI is gaining increasing importance for the noninvasive quantification of organ iron burden. Since transverse relaxation rates depend on iron distribution as well as iron concentration, physiologic and pharmacologic processes that alter iron distribution could change MRI calibration curves. This paper compares the effect of three iron chelators, deferoxamine, deferiprone, and deferasirox on R1 and R2 calibration curves according to two iron loading and chelation strategies. 33 Mongolian gerbils underwent iron loading (iron dextran 500 mg/kg/wk) for 3 weeks followed by 4 weeks of chelation. An additional 56 animals received less aggressive loading (200 mg/kg/week) for 10 weeks, followed by 12 weeks of chelation. R1 and R2 calibration curves were compared to results from 23 iron-loaded animals that had not received chelation. Acute iron loading and chelation biased R1 and R2 from the unchelated reference calibration curves but chelator-specific changes were not observed, suggesting physiologic rather than pharmacologic differences in iron distribution. Long term chelation deferiprone treatment increased liver R1 50% (p<0.01), while long term deferasirox lowered liver R2 30.9% (p<0.0001). The relationship between R1 and R2 and organ iron concentration may depend upon the acuity of iron loading and unloading as well as the iron chelator administered.

INTRODUCTION

Patients with transfusion-dependent anemia, such as thalassemia and sickle cell disease, develop severe iron overload in childhood, leading to endocrine and cardiac complications (1). Lifelong iron chelation is necessary for survival, as well as techniques to monitor therapeutic efficacy. Since iron loading and clearance rates are organ specific, MRI has gained increasing importance in this regard (2–7). Iron increases tissue R2 and R2* and both techniques have been calibrated in the liver and the heart (6,7). A specific R2 analysis procedure (Ferriscan®, Resonance Health, Perth, Australia) was recently FDA-approved for quantification of hepatic iron concentration.

While initial clinical studies have been encouraging, there are valid concerns regarding iron quantification by MRI. Transverse relaxivity appears to be quite sensitive to the size-scale of iron distribution (8–13). Adhering ferritin to 0.8 um diameter synthetic liposomes raises relaxivity 6-fold compared with free ferritin in agar (13). Human siderosomes are similarly sized, accounting for their powerful relaxivity (8). Larger iron “scales” are observed in the reticuloendothelial system and these affect the balance between R2 and R2* relaxivity in the liver (14). Iron chelation is not uniformly effective in all tissue compartments (15,16). This may systematically alter iron storage size and distribution, potentially creating chelator-specific alterations in the relaxivity-iron calibration curves.

Currently deferasirox, deferiprone, and deferoxamine are the only available drugs for iron chelation worldwide. There are many practical barriers to comparing the MRI effects of these drugs in humans, including cost, incomplete drug compliance, intrinsic risk, and variability of biopsy. Many of these shortcomings can be overcome in animal models. Our previous work in unchelated gerbils demonstrated excellent agreement of cardiac and liver R2 with assayed iron concentration, with limits of agreement nearly two-fold better than observed in humans (17).

Thus the goals of this study were to determine whether deferasirox, deferiprone, and deferoxamine systematically disturb R1 and R2-based iron estimates in gerbil liver and heart, relative to unchelated animals. Specifically, we hypothesized that chelation therapy would not systematically shift (bias) the MRI-iron calibration curves but would introduce greater variability in the iron-prediction through perturbations in tissue-iron distribution.

METHODS

All animal studies were performed with approval of the Institutional Animal Care and Use Committee of Children’s Hospital Los Angeles. A total of 112 eight-to-ten week old female Mongolian gerbils (Meriones unguiculatus) were obtained from Charles River Laboratories and housed in the CHLA-accredited animal care facility. Animals were iron loaded and chelated according to 3 different protocols.

Unchelated Protocol

23 animals received weekly, subcutaneous injections of iron dextran (Sigma, St. Louis, MO.) at a dose of 200 mg/kg for a duration ranging from 2–48 weeks. Animals were sacrificed 6 days following their last injection and organs harvested for tissue histology and NMR measurements. Results from 12 of these animals were reported in a previous manuscript (17).

Acute Loading/Unloading

33 animals received three, weekly, subcutaneous injections of iron dextran at 500 mg/kg. Following a 13 day equilibration protocol, chelation was begun with three iron binding compounds: subcutaneously administered deferoxamine (DFO), oral deferiprone (DFP), and oral deferasirox (DSX) for 4 weeks. All chelators were provided by Novartis Pharma AG (Basel, Switzerland). Six animals received no chelation, and the other 27 were randomized to low, medium and high dose chelation therapy; animals were part of a chelator dose-finding study that has been reported elsewhere (15). Deferoxamine in phosphate-buffered-saline (PBS, Invitrogen, Carlsbad, CA) was injected subcutaneously twice daily at a dose of 50,100, and 200 mg/kg/day, five days a week, similar to prior studies (18–20). To avoid the stress of chronic, repeated gavage feeding, deferaprone and deferasirox were homogeneously mixed in 0.1 cc of plain peanut butter (Jif, JM Smucker Company, Orrville, OH) for oral feeding via a 1 cc syringe. Deferasirox was given at single, daily doses 25, 50 and 100 mg/kg, and deferiprone at doses of 125, 250 and 500 mg/kg/day divided into 3 equal doses.

Chronic Loading/Unloading

56 gerbils received 10 weekly injections of subcutaneous iron dextran at 200 mg/kg/day. Following a 13 day equilibration period, 40 animals initiated chelation therapy with subcutaneous deferoxamine (200 mg/kg/day, divided twice daily), deferasirox (100 mg/kg/day, once per day) and deferiprone (375 mg/kg/day divided 3 times per day). 16 animals served as controls; animals were part of a chelator efficacy trial that has been previously reported (16).

Euthanasia, Organ Harvesting and Tissue Histology

Euthanasia was performed with 5% CO₂ according to institutional guidelines. Following sacrifice, the hearts and livers were harvested. The mid-papillary sections of the hearts (~70 mg) and small portions (~130 mg) of the livers were removed, weighed, and sent for quantitative iron determination (Mayo Medical Laboratories, Rochester MN). For reference, human liver biopsy specimens are considered diagnostically adequate if dry weight is greater than 1 mg (21). Tissue dry weight and dry weight iron concentrations were reported by the referral laboratory, allowing calculation of wet-to-dry weight ratios and wet weight iron concentrations. Liver and heart were immersion fixed in 10% formalin, paraffin-embedded and stained with Prussian blue, Masson’s trichrome, and H&E.

NMR Measurements

All studies were performed on a 60 MHz Bruker Minispectrometer. T1 was measured by inversion recovery spin echo imaging with a TR of 4000 ms, TE 2 ms, and 10 inversion times logarithmically spaced between 1 and 1000 ms. T2 was measured by single spin echo technique using a TR of 4000 ms, and 10 echoes logarthmically spaced from 1 to 30 ms (liver) and 1 to 60 ms (heart).

Generation of MRI calibration Curves

R1 and R2 results followed a linear relationship with organ iron concentration except for liver R2 which was nonlinear, similar to humans (6–8). Linear calibration curves were formulated by performing linear regression of iron concentration on NMR relaxation parameter (6–8). Resulting equations were as follows:

$${[\text{Fe}]}_{\text{R}1\text{heart}}=-2.0675+1.9240\ast \text{R}1$$ [1]

$${[\text{Fe}]}_{\text{R}2\text{heart}}=-0.5769+0.02214\ast \text{R}2$$ [2]

$${[\text{Fe}]}_{\text{R}1\text{liver}}=-3.9838+2.8514\ast \text{R}1$$ [3]

To formulate the R2- liver iron calibration curve, the R2-iron relationship was fit to polynomial functions of increasing order to obtain the optimum adjusted r² value (weighted for the degrees of freedom in the model) using the JMP5.1 statistical package (SAS Institute Inc, Cary, NC), yielding

$${[\text{Fe}]}_{\text{R}2\text{liver}}=-3.117922+0.04232\ast \text{R}2+0.0000385{(\text{R}2-279.984)}^2$$ [4]

Residual fitting errors for these models are summarized in the top rows of Tables 1 & 2. In relaxivity-iron measurements, the error scales proportionally to the mean value(22). Geometric confidence intervals are appropriate in this situation and were derived by scaling the predicted calibration curves by the percent uncertainty, represented by the coefficient of variation multiplied by 2.07 (2-tailed T-statistic, alpha 0.05, 22 degrees of freedom).

Table 1.

Bias and Variance of Liver R1, R2 Calibration Curves

Group		Therapy	Bias	Std Dev	Confidence Interval
No Chelation	R1		2.4%	24.4%	[−48%, 53%]
	R2		−1.3%	11.6%	[−25%, 23%]
Acute Chelation	R1	Overall	21.3%	18.2%	[0%, 43%]
		None	14.9%	14.1%	[−18%, 47%]
		DSX	30.8%	24.9%	[−31%, 92%]
		DFP	23.9%	21.1%	[−28%, 76%]
		DFO	17.6%	10.9%	[−9%, 44%]
	R2	Overall	24.7%	19.1%	[2%,47%]
		None	27.1%	14.8%	[−7%, 61 %]
		DSX	28.1%	30.4%	[−46%, 103%]
		DFP	20.5%	19.4%	[−27%, 68%]
		DFO	22.2%	12.8%	[−9%, 54%]
Chronic Chelation	R1	Overall	13.0%	25.8%	[−18%, 47%]
		None	20.3%	21.2%	[−47%, 71%]
		DSX	−9.2%	14.7%	[−48%, 30%]
		DFP	50.0%	18.6%	[5%, 96%]
		DFO	14.1%	16.9%	[−10%, 27%]
	R2	Overall	−0.9%	27.3%	[−35%,32%]
		None	20.8%	23.0%	[−50%, 81%]
		DSX	−30.9%	12.2%	[3%, 66%]
		DFP	5.2%	21.7%	[−48%, 58%]
		DFO	10.3	15.7%	[−7%, 33%]

Bold type indicates statistically significant difference

Table 2.

Bias and Variance of Heart R1, R2 Calibration Curves

Group		Therapy	Bias	Std Dev	Confidence Interval
No Chelation	R1	None	2.4%	30.4%	[−67%, 59%]
	R2	None	0.2%	18.3%	[−38%, 38%]
Acute Chelation	R1	Overall	−27.9%	29.6%	[−63%, 7%]
		None	−31.7%	19.6%	[−14%, 77%]
		DSX	−49.1%	31.7%	[−127%, 29%]
		DFP	−21.2%	23.5%	[−79%, 36%]
		DFO	−8.1%	34.3%	[−92%, 76%]
	R2	Overall	−8.1%	17.3%	[−29%,12%]
		None	−13.9%	17.5%	[−54%, 27%]
		DSX	−16.1%	17.0%	[−58%, 26%]
		DFP	0.2%	12.2%	[−30%, 30%]
		DFO	−0.8%	18.8%	[−47%, 45%]
Chronic Chelation	R1	Overall	4.5%	21.1%	[−20%, 29%]
		None	−3.8%	3.4%	[−12%, 4%]
		DSX	−6.6%	27.4%	[−74%, 60%]
		DFP	23.2%	19.2%	[−24%, 70%]
		DFO	7.0%	13.1%	[−25%, 39%]
	R2	Overall	21.7%	13.7%	[6%, 38%]
		None	16.2%	4.1%	[7%, 26%]
		DSX	19.1%	13.2%	[−13%, 51%]
		DFP	35.3%	15.9%	[−4%, 74%]
		DFO	16.7%	11.1%	[−11%, 44%]

Bold type indicates statistically significant difference

Testing of Calibration Curve in Chelated Animals

Equations [1]–[4] were prospectively evaluated in animals treated using the acute and chronic chelation protocols. Iron estimates by MRI and by chemical assay were compared by Bland-Altman analysis. Bland Altman analysis is appropriate since both measurements contain significant uncertainty. Chemical iron determination has a coefficient of variation of 7% on known standards (Metals Laboratory, Mayo Clinic Rochester, MN). Since MRI uncertainty scales with iron loading (7), all values were expressed as % differences:

$$\text{BA}=({[\text{Fe}]}_{\text{MRI}}-{[\text{Fe}]}_{\text{Chemical}})/(1/2({[\text{Fe}]}_{\text{MRI}}+{[\text{Fe}]}_{\text{Chemical}}))\times 100\%$$

Analysis of variance was performed on the Bland-Altman statistics to determine if there was intergroup variability in bias; post hoc correction was performed using Dunnett’s correction to identify individual group differences. If there were no subgroup differences, results were pooled. Bias in pooled data was evaluated by T-statistic. Differences in variance were evaluated by Fischer’s statistic. A p value of 0.05 was deemed statistically significant.

Reproducibility of NMR methods

The temporal and spatial variability of the NMR methods were assessed by paired measurements in the same organ. For the liver, NMR was performed on samples collected from the left and right lobe. In the heart, a single midpapillary ring was divided into two pieces and measured. Temporal stability of the NMR measurements was performed by repeating the T1 and T2 measurements 30 minutes apart. Lastly, the spatial variability of iron distribution was characterized by measuring the iron in both paired samples. All results are reported using the Bland-Altman statistic.

RESULTS

R1 and R2 from the unchelated animals are demonstrated in Figure 1. Both cardiac R1 and R2 (top panels) rise linearly with iron concentration, having correlation coefficients of 0.87 and 0.91, respectively. Dotted lines represent the 95% confidence intervals for individual animals, not for the regression line and represent the uncertainty for any single measurement. The errors grow proportionally to the mean value (similar to human studies), therefore limits of agreement are reported as percentages rather than absolute values. Using these linear relationships, the error in the MRI-predicted iron concentration is 2.4 ± 30.4 % for R1 and 0.2 ± 18.3 % for R2 (Table 1). This places fairly loose bounds on the predicted cardiac iron concentration, approximately ± 60% for a R1 measurement and ± 37% for a R2 measurement.

Figure 1.

(Top Panels) Cardiac R1 and R2 as a function of wet weight heart iron concentration (in mg/g). R1 and R2 rise linearly with iron concentration. Dotted lines indicate prediction intervals for the regression. (Bottom Panels) Liver R1 and R2 as a function of wet weight liver iron concentration (in mg/g). R1 rises linearly with liver iron concentration. R2 was best fit to a quadratic polynomial. Dotted lines indicate prediction intervals for the regression.

Liver R1 and R2 from the same animals are shown in the lower 2 panels of Figure 1. R1 was linear with iron concentration, having a correlation coefficient of 0.93 R2 was curvilinear, similar to results observed in human liver(6,7), and was well described by a quadratic polynomial. Confidence intervals for predicted iron concentration were 2.4 +/−24.4% for R1 and −1.3% +/− 11.6 for R2 (Table 1). The liver iron prediction error by R2 was less than half the values observed in human studies (6,7), reflecting the homogeneity of the iron loading, increased accuracy of tissue iron determination, improved relaxivity measurements, and absence of chelation.

Figure 2 and Table 1 summarize the calibration curve changes for the liver in response to acute and chronic chelation. Displayed 95% confidence intervals represent the reference range derived from the unchelated animals; changes in Table 1 reflect the bias and variance between treated and untreated animals. Liver R1 was an average of 21.3% higher in acutely treated animals (top left panel), but there was no definitive subgroup effect. Even sham-chelated animals trended toward higher R1 values, suggesting that the more aggressive loading protocol was responsible for the observed changes rather than chelator exposure. Chronically, R1 remained elevated in deferiprone-treated animals (50% ± 18.6%) but not in the other treatment arms (top right panel). Wet-to-dry weight ratio was significantly elevated in the chronically treated deferiprone subgroup, 5.11 ± 0.55 versus 3.85 ± 0.28 for sham-chelated animals, suggesting nonspecific liver toxicity. However, liver histologic appearance was otherwise unremarkable (16).

Figure 2.

Effect of acute and chronic chelation therapy on liver iron calibration curves for animals treated with deferasirox(DSX), deferiprone (DFP), and deferoxamine (DFO). Sham chelated animals are also indicated. Linear fit and prediction intervals derived from the unchelated animals are superimposed. (Top Panels) Liver R1 in acutely and chronically treated animals. (Bottom Panels) Liver R2 in acutely and chronically treated animals.

Liver R2 was positively biased 24.7% in the acutely chelated animals with no drug-specific changes (Figure 2 bottom left panel), again suggesting an effect from aggressive iron loading, rather than chelation. Variability was also increased (19.1% versus 11.6%). Chronic drug therapy eliminated the nonspecific positive bias, but resulted in a –30.9% negative bias in the deferasirox-treated animals (p<0.0001). Ten of the thirteen defersirox-treated animals fell out of the 95% confidence interval. More importantly, R2 values were almost flat with respect to liver iron concentration, making R2 nearly useless in monitoring liver iron for animals chronically treated with deferasirox.

Insight into the deferasirox changes result can be obtained from Figure 3, which compares Prussian blue iron staining from deferasirox, deferiprone, deferoxamine and sham-chelated animals. Three different iron length-scales are evident. The largest of these consists of aggregates of phagocytic cells, having diameters ranging from 50 to 250 um. The second source of iron is the densely stained, comma-shaped, sinusoidal phagocytes, or Kuppfer cells, having dimensions of roughly 10 × 10 × 30 um. The third source of iron is hemosiderin-filled lysomes, called siderosomes, within the hepatocytes themselves. Siderosomes appears as punctate deposits at 1000X magnification and a blue-blush at lower power. Electron microscopy of liver tissue was not performed in these studies but human siderosomes have a mean diameter near 1 um (8). In human iron overload, roughly 75% of the iron pool is stored in hepatocytes and 25% in the reticuloendothelial system (23,24). Animals treated with deferasirox demonstrate excellent clearance from the hepatocyte pool, but no detectable iron removal from the other stores (15,16,25). Deferoxamine and deferiprone removed iron more homogeneously from the different pools.

Figure 3.

Prussian blue staining of liver from representative animals in the chronically chelated group; chelator therapy is indicated in bold type.

The effect of chelation on the heart calibration curves is summarized in Figure 4 and Table 2. Cardiac R1 decreased 27.9% in the acutely loaded/chelated animals (top left panel & Table 2) but this effect was as strong in the sham-chelated animals as in those receiving iron chelation. This negative global bias was not seen in the chronically treated animals (top right panel). There was a trend toward increased R1 in deferiprone treated animals (23.2%± 19.2, p=0.09).

Figure 4.

Effect of acute and chronic chelation therapy on cardiac iron calibration curves. Linear fit and prediction intervals derived from the unchelated animals are superimposed. (Top Panels) Cardiac R1 in acutely and chronically treated animals. (Bottom Panels) Cardiac R2 in acutely and chronically treated animals.

Cardiac R2 demonstrated similar behavior. R2 was 8.1% lower in acutely chelated animals (p = 0.11) but there were no subgroup differences. Chronically, R2 values were 23.7% higher than predicted (p<0.0001). There were again no significant subgroup differences.

The temporal and spatial variability of the MRI and iron measurements are summarized in Table 3. Iron concentration had a coefficient of variation of 11.2% in liver and 15.3% in heart, somewhat lower than observed in human disease (26–28). This error represents spatial variation of iron plus variability from sample digestion and iron assay (at least 7%, John Butz, Mayo Metals Laboratory). The spatial variability of the NMR measurements was of comparable magnitude, suggesting that the NMR measurement error was smaller than the intrinsic sampling error. The temporal variability represents post-mortem changes in tissue water content and diffusion, as well as NMR system drift. Its contribution was minimized by completing all measurements within 20–30 minutes after euthanasia of the animal.

Table 3.

Variability of Tissue Iron and Tissue NMR measurements

Heart
Parameter	Type of Variability	Bland Altman μ ± StD (%)	N
R2	Spatial	5.8 ± 9.3	8
R2	Temporal	1.7 ± 21.9	11
R1	Spatial	9.0 ± 9.9	8
R1	Temporal	2.8 ± 11.9	11
Iron	Spatial	6.8 ± 15.3	22
Liver
R2	Spatial	1.4 ± 5.1	8
R2	Temporal	−2 ± 11.9	11
R1	Spatial	0.8 ± 6.8	8
R1	Temporal	4.7 ± 9.0	11
Iron	Spatial	5.1 ± 11.2	5

DISCUSSION

In clinical practice, MRI-based iron estimation techniques must be able to accurately estimate organ iron in patients treated with different iron chelators. While deferoxamine still remains the most commonly used agent, the use of the oral chelators deferasirox and deferiprone has increased dramatically in the last two years. To date, MRI calibration curves have been based exclusively from patients treated with deferoxamine (6,7,28). We postulated that chelation would increase calibration variability with few systematic biases, however this was not completely true. Two drug-specific shifts were observed in the liver calibration curves and were associated with changes in tissue water content, histology, or iron distribution.

The most clinically relevant observation was the deferasirox-induced shift in the liver R2-iron curve since R2-base iron estimates are being used to guide chelation therapy in iron overloaded patients. The mechanism of this observation is evident from the histology (Figure 3). Deferasirox selectively eliminated iron from the hepatocytes, leading to a predominance of iron stored in larger depots (Kuppfer cells and phagocytic aggregates). These residual deposits produce field defects larger than the scale of water diffusion during R2 measurement, producing static refocusing and decreased signal decay per milligram of iron. The other chelators produced a more balanced depletion of the different pools (iron length-scales) and did not exhibit a similar bias.

It is currently unknown whether deferasirox produces similar R2 changes in humans. Reticuloendothelial and hepatocyte compartments might have equilibrated in this model if given more time. Nonetheless, careful prospective studies in humans are necessary to address whether deferasirox systematically alters MRI calibration curves. Of note, R2* measurements are inherently less scale sensitive than R2 measurements because there is no static refocusing. We recommend that both measurements should be made in deferasirox-treated patients until these calibration uncertainties are resolved.

MRI-based iron estimation techniques must be able to accurately estimate iron in patients having a wide range of iron loading and removal rates. For example, patients with hemochromatosis or thalassemia intermedia adsorb iron at 2–3 times the normal rate (up to 5 mg/day) through constitutive intestinal iron uptake, while thalassemia major patients may accumulate iron at ten-fold the normal rate through chronic transfusion therapy. The present work demonstrates that severity and/or chronicity of iron loading influence MRI-iron calibration curves. Animals who were acutely iron loaded exhibited significant biases in R1 and R2 calibration curves independent of chelation therapy. These changes could arise from any systematic difference in iron distribution, cellular water content, or changes in the magnetic susceptibility of the iron breakdown products. The aggressive iron loading/unloading protocol is inherently more cytotoxic and this may partially explain the differences. The acute model also gives less time for iron to redistribute from its initial reticuloendothelial depots to longer term parenchymal stores.

While analogous iron accumulation and redistribution processes occur in human iron overload, the animal models undoubtedly exaggerate the MRI effects. Human iron accumulation is an order of magnitude slower than our loading protocols. Chelation therapy is also typically gentler. However, extremely aggressive iron chelation is used in patients who present with cardiac dysfunction, often clearing liver iron completely in 6–9 months (4). While relative R2 changes are likely to still be useful in this situation, absolute hepatic iron concentration values should probably be treated with caution because there may have been inadequate time for tissue iron redistribution

R1 measurements have not previously been used for human iron estimation. Previous work in gerbils suggested that R1 measurements might have potential for liver iron quantitation, particularly at very high iron burden (17). In contrast, our present study demonstrates that R1 changes are not selectively determined by iron. For example, deferiprone treated animals had markedly increased R1 values compared with unchelated animals at comparable iron load. R1 changes were accompanied by significant tissue edema which may have been a marker of noninfectious hepatitis. Thus while R1 measurements may be poor markers of iron, it is possible that they might serve as markers of liver toxicity, particularly if co-localized with R2 measurements to control for iron-mediated R1 shortening. This is a relatively unexplored area and warrants further investigation. Unfortunately, accurate R1 measurements are more challenging and time-consuming than R2 measurements in the clinical setting, particularly in iron overloaded patients, limiting the applicability of such techniques.

Limitations

No animal model accurately reflects human iron overload. Iron dextran loading of the gerbil produces initial loading of the reticuloendothelial system in the heart and liver, potentially overemphasizing the contribution of this compartment. In contrast to humans, gerbils also spontaneously eliminate iron at a high rate, predominantly through biliary excretion. If iron chelators interfere with these spontaneous losses, it could potentially cause drug-specific alterations in tissue iron distribution that would not translate to human disease. Gerbils have very high metabolic rates, making it difficult to obtain effective chelation with bolus deferoxamine. This could have been overcome with continuous infusions but there were many practical barriers to doing so.

Another limitation of this study is that the iron loading and chelation protocols represent a heterogeneous mix of iron loading and dosing. The reason for this heterogeneity is that the NMR studies were piggy-backed on chelator studies performed for other indications. However, this shortcoming can also be viewed as a strength, because human patients also have tremendous variation in their transfusion regimens, chelation therapies, and compliance. The error introduced by the broad treatment spectrum is comparable to the error observed in human studies.

Once consequence of chelation is that it tended to “compress” the observed iron scale relative to that observed in the unchelated animals, making it impossible to deduce the “modified” MRI-iron calibration curve of the chelated animals. The sampling of liver iron values from unchelated animals was somewhat sparse and this adds some uncertainty in the predicted iron values in this range. In addition, the quadratic model used to predict liver R2 is only one of many possible curvilinear or piecewise linear models that are consistent with the baseline data. Despite these limitations, predicted iron values from deferasirox treated animals were clearly different from baseline data and from the other treatments. Furthermore, our goal in this manuscript was simply to demonstrate that chelation could change the curve; further studies would be needed to characterize the full nature of these changes.

CONCLUSION

The present study demonstrates that systematic differences in tissue iron distribution introduced by extremes in iron loading and clearance produce large systematic biases in MRI-iron relationships. In particular, intensive deferasirox therapy decreased R2 sensitivity to liver iron. Liver R2 calibrations in humans should be evaluated for potential bias introduced by deferasirox compared with deferoxamine-based norms.