Assessment of in vivo Bone Microarchitecture Changes in a Psoriatic Arthritic Patient Resulting from an Anti-TNFα Treatment

Enrico Soldati (  enrico.soldati@univ-amu.fr ) Aix-Marseille University https://orcid.org/0000-0003-2818-7500 Lucas Esco er Aix-Marseille-University: Aix-Marseille Universite Sophie Gabriel Aix-Marseille-University: Aix-Marseille Universite Jean Pierre Mattei Aix-Marseille-University: Aix-Marseille Universite Serge Cammilleri Aix-Marseille-University: Aix-Marseille Universite David Bendahan Aix-Marseille-University: Aix-Marseille Universite Sandrine Guis Aix-Marseille-University: Aix-Marseille Universite


Introduction
Psoriatic arthritis (PA) is an in ammatory rheumatism associated with psoriasis in which axial and peripheral joints can display an elevated in ammatory status [1]. PA has been initially described by Moll and Wright as a seronegative in ammatory arthritis that occurs most of the time in the presence of psoriasis [2]. It was initially thought to be rare but recent studies indicated that it might occur in up to 30% of patients with psoriasis [3], [4]. The main symptoms are asymmetric oligo-arthritis, polyarthritis, distal interphalangeal joint in ammation, dactylitis, low back pain and enthesis [1], [5]. Structural damage such as bone erosion and bone formation are frequently associated (11). Psoriasis and psoriatic arthritis are characterized by tissue in ltration by activated T cells thereby resulting in an increased TNFa, IL 17 and IL 23 production [5]- [8]. Synovial tissue and entheses are more particularly affected [9]. This pro in ammatory status can be an effective trigger of osteoclasts differentiation and activation through the expression of the receptor activator of nuclear factor kappa B ligand (RANKL) [10].
The systemic bone loss resulting in a reduced bone mineral density (BMD) and the role of TNFα antibodies in this process are a matter of debate in psoriatic arthritis [11]- [14][15]- [17]. So far, these changes have been assessed using dual energy X ray absorptiometry (DXA), which is the gold standard for the diagnosis of osteoporosis [18]. In that respect, BMD values have been reported but bone micro architecture has never been documented as part of this bone alteration process. Interestingly, magnetic resonance imaging (MRI) and more particularly ultra-high eld MRI (UHF MRI) has been reported as a promising tool for the assessment of bone microarchitecture given the high resolution of the corresponding images [19]. Over the last few years, this non-radiating imaging technique has shown promising results regarding the spine and femur trabeculation in osteoporosis [20]- [22]. So far, the corresponding changes in psoriatic arthritis have never been assessed.
The purpose of the present study was to investigate bone trabeculation in a patient with psoriatic arthritis using UHF MRI and to assess changes related to a TNFα antibodies therapeutic strategy.

Subject Recruitment
This study had institutional review board approval and written informed consent was obtained from all the recruited subjects. One PA patient (male, 18 years old, body mass index (BMI) = 14.53 kg/m 2 ) was assessed before and after a one-year Adalimumab treatment. He was naïve of any conventional synthetic
The PA patient was scanned once before treatment and once after one year of treatment. During MRI scanning, the patients' knee was immobilized by sandbags and secured by Velcro straps to avoid involuntary movements.
PET-MRI image fusion MR and CT/PET FNa images [23] were acquired on two different scanners. Given that bones were clearly visible in both CT and MR images, the four bones (femur, tibia, bula, and patella) were used as landmarks for the registration of both images. More speci cally, bones were delineated semiautomatically in each stack of images and linear a ne registrations were computed independently between each bone using FSL-FLIRT [24]. Each local a ne transformation was then merged into a global 3D deformation eld through the implementation described in [25] of the log-euclidean poly-a ne framework proposed by Arsigny et al. [26]. The resulting deformation eld was used to overlay the PET maps on the highly resolved and contrasted 7T MR anatomical images as previously reported [27] ( Fig. 1).

PET-MR Image Analysis
Fused PET-MR images were visually evaluated by an expert (SG) with the aim of identifying and localizing the hypermetabolic regions before and after the treatment. The visual inspection of fused images was crucial in order to identify the regions with hyperintense signals.
Bone volume fraction maps representing the relative volume of bone within each voxel were generated from the GRE images. The initial images were linearly scaled in order to cover the range from 0 (pure bone) to 255 (pure marrow) [28], [29]. On each image, distal femur, proximal tibia and patella were delineated using the Chan-Vese algorithm [30]. The corresponding lled contours were used as masks on which a 10-pixels closing process was applied (2.34 mm) in all directions in order to eliminate the cortical bone ( Fig. 2). Several region of interests (ROI) where identi ed in different locations of the trabecular bone in order to fully investigate the trabecular network.

ROIs selection:
Patella: The rst set of ROIs (ROI1, ROI1a, ROI1b and ROI1c) were located in the patella region and referred respectively to the trabecular space of the whole patella, the upper and lower third of the trabecular region where the quadriceps and patellar tendons are respectively attached and the central third of the patella (Fig. 2).
Distal Femur : ROI2 was located in the distal femur epiphysis as illustrated in Fig. 2.
Proximal Tibia: The nal set of ROIs (ROI3, ROI3a and ROI3b) were positioned in the proximal tibia. ROI3 refers to the trabecular space of the proximal tibia epiphysis. ROI3a represents the trabecular part of the tibia where the medial collateral ligand is attached and ROI3b represents the trabecular part of the tibia where there was no hypermetabolic activity on the basis of the PET FNa signal. (Fig. 2).
These ROIs selection was based on the PET-FNa results. Accordingly they were selected in regions with hyper-intense signals before the Adalimumab treatment and were selected in the same regions after the treatment regardless of the signal intensity.

Bone Microstructure Evaluation
To reduce the computational costs from the 3D ROIs, three 2D centrally located planes were selected for each subject i.e. the image with the highest ROI surface together with the N + 1 and N − 1 images.
ROIs were then binarized using an automatic local thresholding as previously described [31] and three independent metrics were computed. The bone volume fraction (BVF) which refers to the ratio between bone and the total volume, the trabecular thickness (Tb.Th) and spacing (Tb.Sp). Tb.Th and Tb.Sp were extrapolated using iMorph [32] which can generate an aperture map (AM) derived from a distance transformation map. The AM was retrieved from the maximal balls diameter enclosed in the bone (Tb.Th) and in the marrow (Tb.Sp) phases (Fig. 2). Finally the trabecular number (Tb.N) was computed as the ratio between the BVF and the Tb.Th.
Student's T-tests were used in order to assess the morphological parameters differences between the control group and the PA patient before and after the TNF treatment. For each subject, three measurements were obtained for each metric and each ROI. A p-value lower than 0.01 was considered as signi cant.

Standardized Uptake Values
A semi-quantitative analysis of PET images was performed as previously described in order to generate the Standardized Uptake Values (SUV) [23], [33]. SUV were computed as the ratio between the signal intensity within each pixel of the image scaled to the concentration of the total injected radioactivity (3 MBq/Kg). The corresponding results refer the pixel-based metabolic. A SUV of 2.5 or higher is generally considered to be indicative of an "hypermetabolic" region. Finally, mean and maximal values were computed within each ROI.

PET-FNa: Hypermetabolism evolution
The visual inspection of the initial PET image showed intense polyarticular hyperintense signals preferentially involving the knees, the left hip, the right ankle, the elbows, and more moderately the spine, the feet and the hands. As illustrated in Fig. 1, large hyperintensities were observed in the knee. The second PET image recorded after one year of treatment, showed an unequivocal reduction in most of the hypermetabolic regions affecting the joints of the axial and appendicular skeleton and more particularly the knee. The whole set of ROIs showed reduced hyperintensities whereas no more hyperintense signal was visible for ROI2 and ROI3b PET-FNa: SUV results SUV were quanti ed in all the knees ROIs before and after one year of treatment and the corresponding values are indicated in table 2. Before the treatment, SUVmean was abnormal in 5 over 8 ROIs. The abnormal values were concentrated in all the patellar ROIs (2.7 ± 0.1) and ROI3a (2.8). SUVmax averaged over the whole set of ROIs was 3.67 ± 0.41. After the treatment, SUV were no longer larger than 2.5 in almost all the ROIs while the averaged SUVmax was also signi cantly reduced i.e. 2.86 ± 0.86. Large SUV values (i.e. between 1.7 and 2.5) were still visible in all the patella ROIs and ROI3a (Table 2).

MRI microarchitecture
Regarding the MRI-based micro-architecture measurements performed before the treatment, the patient was outside the control range for multiple metrics and multiple localizations (24 out of 32 measurements were statistically different from the controls). However, after one year of treatment the microarchitectural parameters differences between the PA patient and the healthy references were reduced and the parameters were approaching or within the control range (only 9 out of 32 measurements were still statistically different than controls) (table 2).

Patella
Before the treatment and considering the four ROIs delineated in the patellar region, BVF of the patient was always signi cantly lower as compared to controls with a mean difference of 32 ± 19%. The Tb.Th difference was always below 5% (p > 0.01 for all the four ROIs), with a general mean of 0.245 ± 0.010 mm for the controls and 0.244 ± 0.012 mm for the patient. The Tb.Sp difference was statistically signi cant for ROI1, ROI1b and ROI1c but not for ROI1a with the patient having larger trabecular spaces as compared to controls and therefore a negative difference mean of -32 ± 8%. Similar results were found for Tb.N and a signi cant difference was found for ROI1, ROI1b and ROI1c but not for ROI1a with a general mean difference of 27 ± 8%. Following the 12-month of TNF treatment, most of the micro-architecture metrics but Tb.Th reversed to normal values. BVF increased in the four patella's ROIs thereby reducing the differences with controls to a non-signi cant mean value of 2 ± 5%. Similar results were quanti ed for Tb.Sp and Tb.N with a non-signi cant difference with controls for any of the patella's ROIs and a new overall patient mean difference of -9 ± 6% for Tb.Sp and 10 ± 9% for Tb.N. On the contrary, after the treatment, Tb.Th became signi cantly larger with a signi cant difference (up to 8%) with controls and so for ROI1, ROI1b and ROI1c ( Fig. 3 and table 2).

Distal Femur
In the distal femur (ROI2) the difference between the healthy reference and the patient before the treatment was more than 30% for all the parameters (38% for BVF, -33% for Tb.Sp and 32% for Tb.N) except for Tb.Th for which the diffrence was less than 1%.
The image analysis after the treatment still showed increased BVF and Tb.N values for, and reduced Tb.Sp values. The corresponding differences between the patient and the control values were 13%, 14% and − 21% respectively. Similar to the results found in the patella, the Tb.Th increased becoming 3% thicker than controls. The difference between the control and the patient values after the treatment was statistically signi cant (p > 0.01) for none of the micro-architectural parameters evaluated (table 2).

Proximal Tibia
The three ROIs (ROI3, ROI3a and ROI3b) located in the proximal tibia region also showed statistically differences between patient and control values for the whole set of MRI metrics. The only normal value was found for Tb.Th in ROI3a. More particularly, the differences between the patient and the controls were 44 ± 18% for BVF, -34 ± 5% for Tb.Sp, 34 ± 10% for Tb.N and 8 ± 8% for Tb.Th.
After 12 months of TNF treatment, the bone microstructure differences were reduced, although remaining statistically signi cant in most of the cases. For the BVF, the difference was reduced to 26 ± 11% and remained statistically signi cant for ROI3 and ROI3b. The Tb.Th difference was also reduced to 3 ± 6% thereby becoming not statistically signi cant for any of the three tibial ROIs. The Tb.Sp difference slightly decreased to -28 ± 7% but remained statistically signi cant (p < 0.01) for ROI3 and ROI3a but not for ROI3b. The Tb.N difference also decreased to 21 ± 6% but remained statistically signi cant for ROI3 and ROI3a but not for ROI3b (table 2).

Discussion
In the present study, we assessed bone microarchitecture in a PA patient in order to document the potential bone quality changes associated with his in ammatory status. We also assessed the microarchitecture modi cation resulting from a one-year anti-TNF treatment. We mainly found that PET-FNa/MRI showed a largely in amed knee articulation with some speci c hypermetabolic regions in the vicinity of ligament and tendons in the patella, the distal femur, and the proximal tibia. Microarchitectural changes quanti ed using UHF MRI were affecting the whole bone segments and were not localized within the hypermetabolic regions only. After a year of TNF treatment, the combined PET-UHF MRI approach showed highly reduced hypermetabolic regions and an improvement for most of the microarchitectural parameters.
Before the treatment, all the microarchitecture metrics were signi cantly different with respect to the control values and so in at least one ROI. Using HR-pQCT of the distal radius, Kocijan et al. reported similar abnormalities with signi cantly reduced BVF and Tb.N in a group of PA patients [8]. Although previous DXA measurements have been controversial regarding BMD changes in PA patients [11]- [14], our results further support those obtained using a radiating imaging technique and con rm abnormalities of trabecular bone in PA patients so that osteoporotic changes might be expected in PA.
In the eld of rheumatologic in ammatory disorders, our study is the rst to address the bone microarchitecture issue using UHF MRI, although previous studies involving the use of UHF MRI have reported promising results in osteoporosis [20]- [22]. As an example, Chang et al. [20] found abnormal trabecular characteristics including BVF in the distal femur of subjects with fragility fractures whereas the DXA T-score was normal. Of interest, BVF, Tb.Sp and Tb.N were abnormal in the majority (7/8) of ROIs in the present study whereas Tb.Th was abnormal in a limited number (2/8) of ROIs. These results further support those previously reported by Kocijan et al [8] and Chang et al. [20] regarding the larger sensitivity of BVF, Tb.Sp and Tb.N to bone micro-architecture alterations as compared to Tb.Th. In fact, Kocijan et al. [8] reported no difference in Tb.Th between PA patients and healthy controls in distal radii while Chang et al. [20] found normal distal femur Tb.Th in patients with fragility fractures. Trabecular abnormalities detected using UHF MRI were found in all the hypermetabolic regions detected using PET-FNa, showing that microarchitecture deterioration was affecting the whole bone segments. The PET analysis has been shown to re ect bone remodelling and has been used in several studies on osteoporosis [34]- [37]. In our case, PET-FNa allowed us to localize speci c ROIs characterized by elevated hypermetabolic activity before treatment and ROIs presenting partial or full remission after treatment.
After a year of anti-TNF treatment, the trabecular parameters clearly illustrated that the knee of the patient was in clinical remission from his PA status. The trabecular parameters reversal might result from the decreased in ammatory status leading to a reduced osteoclastic bone resorption activity. In PA, Hoff et al. [16] have showed that 24 weeks of In iximab treatment can stop the bone loss. In multiple studies conducted in rheumatoid arthritis (RA) patients, the TNF blocking strategy has been associated with an increase of biologicals markers indicating bone formation and a decrease of those illustrating bone resorption [38]- [40]. In both RA and Ankylosing spondylitis (AS), the e ciency of anti-TNF agents on bone loss has also been con rmed through BMD measurements using DXA (39-41, 42,43). Our PET-FNa/MRI measurements also supported the e ciency of the anti-TNF strategy. In fact, UHF MRI allowed us to assess and quantify the microarchitectural parameters in the hypermetabolic ROIs assessed through the PET-FNa. In our study, UHF MRI showed an almost homogeneous microarchitecture deterioration before treatment and a partial or a complete remission after one year of treatment. These results are also in agreement with those previously reported as a result of bisphosphonates treatment in osteoporotic patients [35], [37].
A few limitations have to be acknowledged in the present study. Although, this preliminary study was conducted in a PA patient we have quanti ed morphological parameters, of several UHF MR images, from 3 different bone segments (patella, distal femur, and proximal tibia) and using 8 different ROIs. Moreover, the results of the PA patient were compared both temporally, i.e. before and after the treatment, and against the control group.
The investigation of bone microarchitecture in patients affected by PA is of interest for a reliable assessment of bone quality, illness risk strati cation and for the follow-up therapeutic strategy. Up to now, PA patients have been mainly treated using CsDMARD, bDMARD and tsDMARD [46] and the improvements of bone microarchitecture have never been assessed. However, the administration of anti-TNF may inhibit the osteoclastic action of bone resorption triggered by the in ammatory response. Moreover, the application of UHF MRI might be of high interest to explore bone microarchitecture in the future and could be applied to investigate some speci c clinical situations.

Declarations
The present study complies with the Declaration of Helsinki, Aix Marseille University ethics committee has approved the research protocol and informed consent has been obtained from the subjects.
All authors have given consent to the publication of the following paper The datasets analyzed during the current study are available from the corresponding author on reasonable request.  The corresponding lled contours were used as masks on which a 10-pixels closing process was applied (2.34 mm) in all directions in order to eliminate the cortical bone The corresponding lled contours were used as masks on which a 10-pixels closing process was applied (2.34 mm) in all directions in order to eliminate the cortical bone Figure 3 On the contrary, after the treatment, Tb.Th became signi cantly larger with a signi cant difference (up to 8%) with controls and so for ROI1, ROI1b and ROI1c Figure 3 On the contrary, after the treatment, Tb.Th became signi cantly larger with a signi cant difference (up to 8%) with controls and so for ROI1, ROI1b and ROI1c

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.