Molecular test of Paget's disease of bone in families not linked to SQSTM1 gene mutations

Abstract

Purpose

Paget's disease of bone (PDB) is a focal metabolic bone disorder characterized by an increased bone remodeling. Fifteen to 40 % of PDB patients have a familial form with an autosomal dominant inheritance. Disease-causing mutations of the SQSTM1 gene have been linked to PDB in about 40 % of families whereas genes linked to the remaining families are unknown. Several single nucleotide polymorphisms (SNPs) have been associated with PDB in unrelated patient non-carriers of a SQSTM1 mutation. The current clinical practice guidelines still recommend the measure of serum total alkaline phosphatase (sALP) for PDB screening. In unrelated individual non-carriers of SQSTM1 mutations, we previously developed a genetic test combining male sex with five genetic markers (rs499345, rs5742915, rs2458413, rs3018362, rs2234968), giving rise to an area under the curve (AUC) for PDB phenotype of 0.73 (0.69; 0.77). A combination of male sex with total calcium corrected for albumin and Procollagen type I N-terminal propeptide (P1NP), had an AUC of 0.82 (0.73; 0.92). Combining both genetic and biochemical tests increased the AUC to 0.89 (0.83; 0.95).

Objective

This study aimed at estimating the performance of our previous test of PDB, in families not linked to SQSTM1 mutations with disease-causing genes yet unknown, and at developing a new algorithm if the performance is not satisfactory.

Methods

We genotyped the five SNPs cited above, and measured calcium corrected for albumin and P1NP in 181 relatives, with PDB or not, from 19 PDB families not linked to SQSTM1 mutations. Bivariate and multivariate logistic regression models including male sex were fitted to search for a molecular test that could best detect PDB in these families. A receiving operating characteristics analysis was done to establish a cut-off point for continuous variables.

Results

Logistic regression estimates of our previous molecular test gave rise to a high sensitivity of 78 %, 97 % and 88 % for the genetic, biochemical, and combined test but the specificity was very low, 35 %, 11 % and 21 %, respectively. This poor specificity persisted even when the cut-off point was changed. We then generated in these families, new logistic regression estimates but on the same parameters as mentioned above, giving rise to an AUC of 0.65 (0.55; 0.75) for the genetic test, of 0.84 (0.74; 0.94) for the biochemical test, and 0.89 (0.82; 0.96) for the combination test, the latter having a sensitivity of 96 % and specificity of 57 %. By comparison serum P1NP alone gave rise to an AUC of 0.84 (0.73; 0.94), with a sensitivity of 71 % and a specificity of 79 %.

Conclusion

In PDB families not linked to SQSTM1 mutations, the estimates of our previous molecular test gave rise to a poor specificity. Using new estimates, the biochemical and combined tests have similar predictive abilities than our former test. Serum P1NP is a bone marker of interest for the screening for PDB in families not linked to SQSTM1 mutations.

1. Introduction

Paget's disease of bone (PDB) is a late-onset disease characterized by a highly variable prevalence, from 5.4 % in the UK to 0.00028 % in Japan (Corral-Gudino et al., 2013), the latter increasing with age. But in the UK, the incidence of PDB was reported to decline by 60 % in the recent years (Cook et al., 2021; Cooper et al., 2006), as well as a decline in the clinical severity of PDB was also reported (Tan and Ralston, 2014). This is likely related to changes in environmental factors which contributed to decrease the expressivity of PDB, such as an effect on macro-autophagy (Hocking et al., 2012).

PDB remains the second most common metabolic bone disease after osteoporosis, affecting slightly more men than women. PDB is characterized by focal abnormal bone remodeling, due to an increased bone resorption followed by an increased and disorganized new bone formation leading to an anarchic bone structure (Galson and Roodman, 2014; Roodman and Windle, 2005).

PDB has a strong genetic component, since the familial form is present in 15–40 % of patients, with autosomal dominant inheritance through incomplete penetrance (Laurin et al., 2001; Siris, 1994; Michou et al., 2006). SQSTM1 gene mutations are associated with 10 % of cases with PDB (Laurin et al., 2002; Morissette et al., 2006), and associated with 40 % of familial forms of PDB, whereas genes linked to the remaining families are unknown. Several single nucleotide polymorphisms (SNPs) associated with PDB are identified in patient non-carriers of a SQSTM1 mutation. Among the SNPs associated with PDB, we find common variants of the TNFSFR11A, VCP, OPTN and DKK1 genes. Rare genetic variants in the UCMA/GRP, OPTN, DKK1 and TM7SF4 genes have also been associated with PDB (Albagha et al., 2010; Albagha et al., 2011; Beauregard et al., 2014; Beauregard et al., 2013; Michou et al., 2012; Silva et al., 2018a; Silva et al., 2018b).

Patients with PDB are frequently asymptomatic, the diagnosis therefore is often incidental. Only 10 to 30 % of patients will be symptomatic (Tan and Ralston, 2014; Roodman and Windle, 2005). Signs and symptoms that suggest PDB include bone pain, deformity, fracture, neurological abnormalities, such as headache, deafness, abnormal gait and osteoarthritis, cardiovascular complications, and osteosarcoma, the latter being the most feared (Tan and Ralston, 2014; van Staa et al., 2002; Wermers et al., 2008; Saraux et al., 2007; Seton et al., 2003; Merlotti et al., 2005). The treatment of choice in PDB is bisphosphonates, most notably zoledronic acid, which has a high level of effectiveness (Reid et al., 2005; Corral-Gudino et al., 2017). A single zoledronic acid infusion induces a response that lasts for up to 10 years (Reid et al., 2013; Reid et al., 2011).

In Canada, excluding PDB is recommended before prescribing bone anabolic agents since PDB is an absolute contraindication to PTH analogs, which are increasingly used for osteoporosis treatment. PTH analogs were suspected to increase the risk of osteosarcoma in rat toxicologic studies (Vahle et al., 2002). This poses a challenge since PDB affects 2–6 % of the Caucasian aging population and is frequently asymptomatic.

Various methods have been used in clinical practice to detect PDB. Measurement of serum total alkaline phosphatase levels, a whole-body bone scintigraphy, followed by radiographs, particularly of the skull and enlarged pelvis including the lower lumbar spine and the upper third femurs, are recommended (Singer et al., 2014). Scintigraphy remains the most sensitive method for diagnosing PDB, however, it is expensive and invasive (Singer et al., 2014). To date, there is no reliable, non-invasive, and cost-effective test to detect an asymptomatic PDB.

The current clinical practice guidelines recommend the measure of serum total serum alkaline phosphatase (sALP) for PDB screening. sALP may often be normal in PDB, resulting in false negatives, especially in monostotic or metabolically inactive disease (Ralston, 2013). Therefore, we have previously identified genetic and biochemical markers for PDB screening (Guay-Belanger et al., 2016). In American, French and French-Canadian patients with PDB, non-carriers of a SQSTM1 mutation, a combination of five SNPs (rs499345 (locus 1p13, genes EPS8L3/CSF-1), rs5742915 (15q24, PML), rs2458413 (8q22, TM7SF4), rs3018362 (18q21, RPL17P14) and rs2234968 (10p13, OPTN) over 35 SNPs tested due to their association with PDB in the literature), including male sex, gave rise to an area under the curve (AUC) of 0.731 (0.688; 0.773), with a sensitivity of 82 % and a specificity of 51 %. A combination of serum calcium corrected for albumin and serum P1NP, adjusted for sex, has an AUC of 0.822 (0.726; 0.918), with a sensitivity of 82 % and a specificity of 49 %, whereas other biochemical tests such as 25-OH vitamin D, interleukin-6, parathyroid hormone, C-telopeptide, N-mid osteocalcin, high-sensitivity C-reactive protein, receptor activator of nuclear factor κ-B ligand, and osteoprotegerin were not retained in the final model due to lack of association with PDB or low AUC. A combination of the five SNPs and serum calcium corrected for albumin and serum P1NP increased the AUC to 0.892 (0.833; 0.951), with a sensitivity of 89 % and a specificity of 54 % (Guay-Belanger et al., 2016). Finally, we proposed a two-step algorithm, consisting first of a screen for SQSTM1 mutations, followed by a genetic test or a combined genetic and biochemical test. The genetic algorithm correctly identified 83.6 % of patients with PDB, with a sensitivity of 84 % and a specificity of 51 %, while the algorithm integrating both genetic and biochemical markers identified 94 % patients with PDB, with a sensitivity of 94 % and a specificity of 54 % (Guay-Belanger et al., 2016).

This study aimed at establishing the performance of our previous test as published by Guay-Belanger et al. (2016) in French-Canadian families not linked to SQSTM1 gene mutations, in whom disease-causing genes are yet unknown, and at developing a new algorithm if the performance of the previous test is not satisfactory.

2. Methods

The study was approved by the ethics committee of the CHU de Québec-Université Laval (IRB number 2020-4729). All study participants clinical data and biological sample (DNA, serum) originated from French-Canadian families stored in our biobank on heritable musculoskeletal diseases in Quebec City (Canada). One hundred eighty-one relatives, from 19 families not linked to SQSTM1 mutations, were included in the study, among which there were 45 relatives with PDB (24.9 %) and 136 relatives without PDB (75.1 %). The median number of affected relatives who participate to this study was 2 (interquartile range 2–3) and the median number of unaffected relatives was 5 (interquartile range 3–7). The formal analysis of pedigrees was consistent with an autosomal pattern of inheritance in five families. In the 14 remaining families, an autosomal dominant or recessive pattern of inheritance remained possible, as all affected relatives were siblings and no PDB was reported in their parents, now deceased. We collected clinical data on sex and age at PDB diagnosis. Measurement of sALP, radiographs, and a whole bone scintigraphy were already available for this patient cohort, collected in our registry and completed with clinical data available from electronic medical records at CHU de Quebec-Université Laval. Radiographs include those of the skull and pelvis for all participants, and additional radiographs of the lower limbs, femur, tibia, hemithorax, hand, wrist, forearm, elbow, feet, spine, knee, collarbones, shoulder, scapula, humerus, or ankle were performed only in bones with increased uptake on bone scintigraphy. The percentage of skeletal extension of PDB was calculated by the Renier's index (Renier et al., 1995). Clinical diagnosis of PDB relied on: 1) an abnormal whole-body bone scintigraphy, 2) typical signs of PDB on the bone radiographs, and/or 3) an increase in sALP level, as published (Laurin et al., 2001).

We genotyped the five SNPs cited above (rs499345, rs5742915, rs2458413, rs3018362 and rs2234968) in 168 of 181 relatives in total, after excluding 13 participants for whom no DNA sample was available, 35 of 45 relatives with PDB (20.8 %) and 133 of 136 relatives without PDB (79.2 %), from our 19 families not linked to SQSTM1 mutations. Genotyping of SNPs was done using the Sequenom MassARRAY SNP Multiplex Technology, completed by Sanger sequencing for missing data at the Plateforme de Séquençage et de Génotypage des Génomes du Centre de Recherche du CHU de Québec, as published (Guay-Belanger et al., 2016).

We also measured serum calcium corrected for albumin and serum P1NP, in 143 of 181 relatives with PDB (n = 29 of 45) or not (n = 114 of 136), from our 19 families with PDB not linked to SQSTM1 mutations, after excluding missing values due to lack of serum sample for 38 participants. Calcium and albumin were assayed using commercial kits with the AU5800 analyzer (Beckman Coulter Inc., California, USA). Procollagen type 1 amino-terminal propeptide (P1NP) was measured using commercial Roche Diagnostics kits with the Cobas e411 system (Hoffman's division La Roche Ltd.; Laval, Canada). All these assays were performed at the CHUM (Montréal, Canada). The serum calcium was corrected for albumin using the standard formula (corrected calcium (mmol/L) = total calcium (mmol/L) + 0.02 × [40 − albumin (g/L)]).

2.1. Statistical analyses

First, by using the pre-established equations, as follows:

• (1)Odds of disease = Exp [−2.8535 + (sex ∗ 1.2590) + (rs499345 ∗ 0.5316) + (rs5742915 ∗ 0.6064) + (rs2458413 ∗ 0.7487) + (rs3018362 ∗ 0.7037) + (rs2234968 ∗ 0.5751)];
• (2)Odds of disease = Exp [−27.6491 + (sex ∗ 0.6241) + (Calcium ∗ 0.0106) + (P1NP ∗ 0.0184)];
• (3)Odds of disease = Exp [−30.9036 + (sex ∗ 0.5957) + (rs499345 ∗ 0.4824) + (rs5742915 ∗ 1.3519) + (rs2458413 ∗ 0.3675) + (rs3018362 ∗ 1.6467) + (rs2234968 ∗ 0.4279) + (calcium ∗ 0.0106) + (P1NP ∗ 0.0202)])

and the previously determined cut-off point of 0.33 for the predicted probability obtained by logistic model on genetic factors; of 0.07 for the predicted probability obtained by logistic model on biochemical measures; and 0.05 for the predicted probability obtained by logistic model on genetic and biochemical data (Guay-Belanger et al., 2016). For all these models, we computed area under the receiving operating characteristics (ROC) curves and 95 % CIs, and calculated intrinsic characteristics, including sensitivity, specificity, as well as extrinsic characteristics, including positive and negative predicting values (PPV and NPV) with 95 % CIs, for the genetic, biochemical, and combined test, using corresponding cut-offs.

Then, new bivariate and multivariate logistic regression models (generalized mixed model – SAS GLIMMIX procedure, with family [belonging to the same family] as a random intercept) adjusted for sex were also fitted to search biochemical and genetic markers that could best detect PDB. A logistic model was also applied for the combination of SNPs and biomarkers. The predicted probability of PDB, estimated by these models, was calculated for each participant. The ROC curve analysis (using a sensitivity maximization with a specificity of at least 50 %) was applied to establish a cut-off point of the predicted probability estimated by logistic regression model. The area under the ROC curves and 95 % CIs were estimated for all these models using DeLong et al.'s approach available in SAS. For all genetic, biochemical, and combined models, we presented intrinsic characteristics, including sensitivity, specificity, as well as extrinsic characteristics, including positive and negative predicting values (PPV and NPV) with 95 % CIs. Descriptive statistics are presented with a mean ± standard deviation, frequency, and percentages. All statistical analyses were performed at the CHU de Québec-Université Laval research centre, using SAS 9.4.

3. Results

3.1. Description of the study population

Our study population consisted of 181 relatives from 19 families with PDB, 94 men (51.9 %) and 87 women (48.1 %). Forty-five (45) family members had a diagnosis of PDB, and among them, 23 men (50 %). In family members with PDB, their mean age at PDB diagnosis was 62.0 ± 11.0 years, ranging from 39 to 84 years old. Their mean for total serum alkaline phosphatase (sALP) was 1.9 ± 1.5 times the upper limit of normal. Twenty patients (43.48 %) had a monostotic disease. The average number of bone sites affected by PDB was 3.0 ± 2.0 bones, ranging from 1 to 12. The mean percentage of bone area affected by PDB as determined by the Renier's index was 10.1 % ± 7.5 %, varying from 0.65 % to 33.70 %. For the genetic test, we studied 168 of 181 participants for whom a DNA sample was available (81 men (48.2 %) and 87 women (51.8 %)). The sub-group of participants for the biochemical test consisted of 143 of 181 participants with PDB (n = 29 of 45) or not (n = 114 of 136), for whom a serum sample was available.

For the genetic, biochemical, and combined test, logistic regression estimates of our previous molecular test as published by Guay-Belanger et al. (2016) gave rise to a high sensitivity, however the specificity was very low (Table 1). For the genetic test, logistic regression estimates of our previous test indeed gave rise to a sensitivity of 78 % and a specificity of 35 %. A cut-off point of 0.33 was used. Of the 126 participants without PDB, 44 participants (26.5 % of total participants) had a predicted probability below the 0.33 cut-off point and 82 participants (50.96 %) had a predicted probability >0.33. Of the 40 participants with PDB, 9 participants (5.4 %) had a predicted probability <0.33 and 31 participants (18.7 %) had a predicted probability >0.33 (Table 1). For the biochemical test, logistic regression estimates of our previous test (Guay-Belanger et al., 2016), using a cut-off point of 0.07, gave rise to a sensitivity of 97 % and a specificity of 11 % (Table 1). For the combined test, logistic regression estimates of our previous test (Guay-Belanger et al., 2016), using a cut-off point of 0.05, gave rise to a sensitivity of 89 % and a specificity of 21 %. Of the 106 participants without PDB, 22 participants (16.7 % of total participants) had a predicted probability <0.05 and 84 participants (63.6 %) had a predicted probability >0.05. Of the 26 participants with PDB, 3 participants (2.3 %) had a predicted probability <0.05 and 23 participants (17.4 %) had a predicted probability >0.05 (Table 1).

Table 1.

Characteristics of the previous and the new molecular tests.

	Previous molecular testa			New molecular test
	Genetic test	Biochemical test	Combined test	Genetic test	Biochemical test	Combined test
Cut-off point of the predicted probability	>0.33	>0.07	>0.05	>0.23	>0.16	>0.09
True positive, n	31	28	23	28	24	25
True negative, n	44	13	22	63	91	60
False positive, n	82	101	84	63	23	46
False negative, n	9	1	3	12	5	1
Sensitivity (95 % CI), %	77.5 (61.6;89.2)	96.6 (82.2;99.9)	88.5 (70.0;97.6)	70.0 (53.5;83.4)	83.0 (64.2;94.2)	96.0 (80.4;99.9)
Specificity (95 % CI), %	34.9 (26.7;43.9)	11.4 (6.2;18.7)	20.8 (13.5;29.7)	50.0 (41.0;59.0)	80.0 (71.3;86.8)	57.0 (46.6;66.2)
Positive predictive value (95 % CI), %	27.4 (19.5;36.6)	21.7 (14.9;29.8)	21.5 (14.1;30.5)	31.0 (21.5;41.3)	51.0 (36.1;65.9)	35.0 (24.2;47.5)
Negative predictive value (95 % CI), %	83.0 (70.2;91.9)	92.9 (66.1;99.8)	88.0 (68.8;97.5)	84.0 (73.7;91.5)	95.0 (88.3;98.3)	98.0 (91.2;100)

95 % CI = 95 % confidence interval.

^aAs published in Guay-Belanger et al. (2016).

Given the high sensitivity but the very low specificity of our former equations using the previously determined cut-off points for the three tests (Table 1) as well as higher or lower cut-off points for each test (Table 2), we then generated in these families, new logistic regression estimates but on the same parameters, as mentioned above.

Table 2.

Characteristics of the previous molecular test with adjustments for lower and/or higher cut-off points.

	Previous molecular testa
	Genetic test			Biochemical test			Combined test
Cut-off point of the predicted probability and adjustments for lower and/or higher cut-off points	>0.33	>0.23	>0.43	>0.07	>0.12	>0.15	>0.05	>0.10	>0.15
True positive, n	31	37	29	28	22	20	23	20	20
True negative, n	44	9	57	13	41	57	22	44	54
False positive, n	82	117	69	101	73	57	84	62	52
False negative, n	9	3	11	1	7	9	3	6	6
Sensitivity (95 % CI), %	77.5 (61.6;89.2)	92.5 (79.6;98.4)	72.5 (56.1;85.4)	96.6 (82.2;99.9)	75.9 (56.5;89.7)	69.0 (49.2;84.7)	88.5 (70.0;97.6)	76.9 (56.4;91.0)	76.9 (56.4;91.0)
Specificity (95 % CI), %	34.9 (26.7;43.9)	7.1 (3.3;13.1)	45.2 (36.4;54.4)	11.4 (6.2;18.7)	36.0 (27.2;45.5)	50.0 (40.5;37.2)	20.8 (13.5;29.7)	41.5 (32.0;51.5)	50.9 (41.1;60.8)
Positive predictive value (95 % CI), %	27.4 (19.5;36.6)	24.0 (17.5;31.6)	29.6 (20.8;40.0)	21.7 (14.9;29.8)	23.2 (15.1;32.9)	26.0 (16.6;37.2)	21.5 (14.1;30.5)	24.4 (15.6;35.1)	27.8 (17.9;39.6)
Negative predictive value (95 % CI), %	83.0 (70.2;91.9)	75.0 (42.8;94.5)	83.8 (72.9;91.6)	92.9 (66.1;99.8)	85.4 (72.2;93.9)	86.4 (75.7;93.6)	88.0 (68.8;97.5)	88.0 (76.0;95.5)	90.0 (79.5;96.2)

95 % CI = 95 % confidence interval.

^aAs published in Guay-Belanger et al. (2016).

The new genetic test gave rise to an AUC of 0.65 (0.55; 0.75) (Fig. 1A). Using a cut-off point of 0.23, this five-genetic-marker combination, including male sex, had a sensitivity of 70 %, a specificity of 50 %. PPV and NPV were respectively at 31 % and 84 %. The new biochemical test gave rise to an AUC of 0.84 (0.74; 0.94) (Fig. 1B). Using a cut-off point of 0.16, the two-biomarker combination, consisting in serum calcium corrected for albumin and serum P1NP, including male sex, had a sensitivity of 83 %, a specificity of 80 %. PPV and NPV were respectively at 51 % and 95 %. The new combination test, consisting of both genetic and biochemical markers, gave rise to an AUC of 0.89 (0.82; 0.96) (Fig. 1C). Using a cut-off point of 0.09 for the combined score, this combination, including male sex, had a sensitivity of 96 %, a specificity of 57 %. PPV and NPV were respectively at 35 % and 98 % (Table 1).

Fig. 1.

Area under the receiving operating characteristics curves of the screening for Paget's disease of bone in families not linked to SQSTM1 gene mutations: A) new genetic test, B) new biochemical test, and C) new combined test.

We also found out that serum P1NP alone gave rise to an AUC of 0.84 (0.74; 0.94). Using a cut-off point of 0.15, serum P1NP alone had a sensitivity of 83 %, a specificity of 77 %, and a PPV and NPV respectively of 48 % and 95 % (Fig. 2).

Fig. 2.

Area under the receiving operating characteristics curve of the Procollagen type I N-terminal propeptide alone.

4. Discussion

In relatives of families with PDB not linked to any SQSTM1 mutations, logistic regression estimates of our previous genetic tests combining either five genetic markers, serum calcium corrected for albumin and serum P1NP, or both (all also including sex) as published by Guay-Belanger et al. (2016), had moderate to high sensitivity but low specificity at the disease probability cut-off points determined previously. Poor discrimination of relatives with and without PDB irrespective of the cut-off point led us to generate in these families new logistic regression estimates. The new genetic test had a sensitivity of 70 % and a specificity of 50 %, and the new biochemical test had a sensitivity of 83 % and a specificity of 80 %. The new combination test increased the AUC to 0.89 (0.82; 0.96) and had a sensitivity of 96 % and a specificity of 57 %, whereas serum P1NP alone gave rise to an AUC of 0.84 (0.73; 0.94), having a sensitivity of 71 % and a specificity of 79 %.

After generating new logistic regression estimates, we found that our new genetic test was less likely to predict PDB in these families than our previous test (Guay-Belanger et al., 2016). This finding may be explained by the fact that we studied related individuals, who may share at risk alleles with their affected relatives. The biochemical and combined tests have similar predictive abilities than our former test. We also noticed that serum P1NP alone has very interesting results in terms of AUC, sensitivity, and specificity. Our results then suggest that the screening for PDB using serum P1NP could be an interesting tool to improve the diagnosis of PDB in families not linked to SQSTM1 mutations, with disease causing genes being yet unknown. Serum P1NP has a sensitivity of 77–100 % but is not currently widely accessible for clinical practice, due to its cost (Al Nofal et al., 2015). Although the P1NP is not widely available in clinical practice yet, it will be interesting to include P1NP in our molecular tests in the future, as a valuable tool for the screening of PDB. Scintigraphy remains the most sensitive method for detecting PDB, being the gold standard with a sensitivity of 97 % to 98 % (Michou and Orcel, 2016). It is however costly and more invasive than blood tests. In a personalized medicine era, developing innovative molecular testing for PDB screening is a promising way to decrease the systematic need for whole-body bone scan. However, studying bone biomarkers which reflect PDB activity may be challenging considering the declining severity of PDB, with a higher proportion of monostotic PDB. Furthermore, in familial forms not linked to SQSTM1 gene, the SNPs associated with PDB in former studies, mostly related to the multifactorial genetic component of PDB, has a limited added value, as demonstrated by our results. Another molecular test has been developed for PDB management, relying on seven SNPs (rs10494112, rs4294134, rs2458413, rs1561570, rs10498635, rs5742915, and rs3018362) associated with PDB in the GWAS in an additive manner. This cumulative risk allele score, which was not developed in a PDB screening purpose like ours, was reported to predict the extent and severity of PDB with a sensitivity of 70 % and a specificity of 55 % (Albagha et al., 2013). These seven SNPs have been included in our former molecular test (Guay-Belanger et al., 2016). The SNP rs10498635 was removed from the former study because of departure from Hardy-Weinberg equilibrium, and the rs10494112 was removed because of complete linkage disequilibrium with other studied SNPs (Guay-Belanger et al., 2016). Three SNPs, rs2458413, rs5742915 and rs3018362 of our genetic test were also part of the cumulative risk allele score to predict the severity of PDB.

Our study has some limitations. Only 19 families with PDB not linked to SQSTM1 mutations were available to this study. Furthermore, the presence of missing or incomplete data for some participants from our registry and missing genetic and biochemical data from our biobank have limited the size of our sample. The specificities of the genetic, biochemical, and combined tests remained low, however better sensitivity, at the expense of specificity, has been prioritized to identify the maximum number of patients with PDB.

In the future, testing the intrinsic properties of this molecular test at detecting PDB in larger cohorts would be of interest, as well as evaluating its performance at predicting disease severity, in particular the risk of neoplastic degeneration occurring on pagetic bones. Searching for a combination including additional SNPs known to be associated with PDB in the literature could possibly lead to better sensitivity and specificity for the detection of PDB in non-familial cases, which are currently the most frequent in the clinical practice.

5. Conclusion

In conclusion, in families with PDB not linked to SQSTM1 mutations, the logistic regression estimates of our previous molecular test of PDB gave rise to a high sensitivity but poor specificity. Using new logistic regression estimates, our genetic test was less likely to predict PDB phenotype in relatives of these families than our previous test in unrelated participants, but the biochemical and combined tests have similar predictive abilities than our former test. Serum P1NP alone appears to be promising in screening for PDB in families not linked to SQSTM1 mutations.

Declaration of competing interest

Yang You has no competing interest in relation to this article.

David Simonyan has no competing interest in relation to this article.

Alexandre Bureau has no competing interest in relation to this article.

Edith Gagnon has no competing interest in relation to this article.

Caroline Albert has no competing interest in relation to this article.

Jason R Guertin has no competing interest in relation to this article.

Jean-Eric Tarride has received research grants from Assurex/Myriad, Edwards LifeSciences, and Boehringer Ingelheim; and consultant payments from Amgen, Bayer, Evidera, Analytica Laser International, Lilly, Merck, Novartis, Novo Nordisk, Roche and Pfizer outside of this work.

Jacques P. Brown has received research support from Mereo BioPharma, Radius Health, and Servier; has served as a consultant for Amgen, Gilead, Paladin, Pfizer, Servier and Ultragenyx; and has served on speakers' bureau for Amgen, all outside the scope of this manuscript.

Laetitia Michou has received honoraria for a conference from Roche, Janssen, Abbvie, Amgen; has served as consultant on Advisory Boards of Pfizer, Roche, Amgen, outside the scope of this paper.

Data availability

Data will be made available on request.