Introduction
Osteogenesis imperfecta (OI) is a rare congenital musculoskeletal disorder caused by mutations in ~ 20 genes related to type I collagen synthesis, with impact on osteoblast differentiation and mineralization in bone [
1,
2]. OI patients present with low bone density, high fracture rates, long bone deformity, scoliosis and a wide array of other symptoms [
3,
4]. Current treatments, including bone density enhancement and orthopedic corrections [
5], do not fundamentally cure the condition, causing considerable burdens on affected individuals and society.
Traditionally, OI was clinically grouped into four subtypes (I, II, III and IV) based on the Sillence classification, with type I being mildest, followed by type IV and then type III. Type II is perinatally lethal and thus most serious [
6]. The modified classification includes the original four subtypes and a new type V that has unique clinical phenotypes and is uniquely caused by a single-point mutation in the 5’-UTR of
IFITM5 (c. -14C > T) [
7,
8]. With rapid technical advances in the past decade, OI is also classified genetically. Patients with
COL1A1/2 and
IFITM5 mutations remain categorized according to the Sillence scheme, while those with mutations in the other 17 genes are subtyped from VI (OMIM #613982) to XXII (OMIM #619795) [
4].
Scoliosis, which affects physical mobilities and cardiopulmonary functions, is a form of lateral deformity of the spine (defined as Cobb angle > 10°), categorized into idiopathic (i.e. unknown causes), congenital and neuromuscular subtypes [
9]. Scoliosis is commonly found in OI patients [
10], with an estimated prevalence of around 50% [
11,
12]. Scoliosis in OI is progressive, with an estimated Cobb angle increase of 2.3°–2.6° per year [
11‐
14]. The causes of OI scoliosis are not clear. However, genetics [
15], age, gender, drug treatment [
16], bone density [
17], Sillence classification, and deformities in the limbs or joints [
18] are implicated as potential risk factors. Previous efforts studied the risk factors associated with OI scoliosis and suggested potential treatment strategies. Unfortunately, these studies had either small sample sizes [
16] or incomplete genotypes [
12].
To delineate the relationship between OI scoliosis and potential risk factors, we retrieved the medical records of all consecutive OI patients from 2014 to 2022 seeking treatments at our hospital. Genetic testing results were included where possible, with affected genes covering
COL1A1/A2,
IFITM5,
WNT1,
SERPINF1,
FKBP10, etc. Genetic inheritance (AD/AR) and mutation types (qualitative or quantitative) were also documented. We also retrieved information of skeletal maturity, bone density, drug history, Sillence subtypes, and conditions in the limbs. Based on the Cobb angle of the major curves, we stratified the patients into four severity grades: non-scoliotic, mild, moderate and severe [
19]. We then performed univariate and multivariate analyses between the independent variables and the severity outcomes. We estimated the progression rates based on longitudinal radiographs and conducted multiple linear regression to identify associated factors.
Methods
Samples and materials
Records of all patients diagnosed with osteogenesis imperfecta (OI) in our hospital from August 2014 to November 2022 were retrieved for the current study (n = 308). Eighteen patients without radiographs of the spine were excluded. For each of the remaining 290 patients with spine radiographs, sitting or standing Cobb angles at the thoracic, thoracic-lumbar (TL), and lumbar regions were measured. Each patient may thus have up to three curves, of which the one with the maximum Cobb angle was designated the major curve. All Cobb angles were measured by three experienced pediatric orthopedic surgeons (YPZ, YZL and DLLL). Each patient had one or multiple spinal radiograph follow-ups. Data from the follow-up with maximum major curve Cobb angle was used to grade scoliosis severity. Patients with Cobb angle below 10° were considered non-scoliotic, while those with preoperative Cobb angles between 10°–25°, 25°–50° and > 50° were considered mild, moderate and severe scoliosis, respectively.
Drug treatment history, including the dates and types of anti-osteoporotic agents, was retrieved. Patients were contacted for confirmation of first date of drug treatment and the menarches for female patients. The BMD data were retrieved from the Discovery DXA system (Hologic Inc., Massachusetts) at our hospital. The total hip and lumber regions (L1–L4) BMDs were used. Weight and height corresponding to each BMD measurement were also collected.
The patients were clinically categorized based on the radiographic features, BMD reports and drug treatment histories according to the criteria of modified Sillence classification [
7], where patients were labelled as types I–V. At least two of our pediatric orthopedic surgeons (YPZ, YZL, DLLL, JWW and LF) were involved in independently rating each patient. In case of ambiguity, three or more physicians were invited to rate for a final consensus. MKTT reviewed and approved the final subtyping results.
Genetic sequencing
Targeted amplicon sequencing was performed on 221 out of the 290 patients. Nineteen OI causative genes (including
COL1A1, COL1A2, IFITM5, SERPINF1, CRTAP, P3H1, PPIB, SERPINH1, FKBP10, BMP1, SP7, TMEM38B, WNT1, CREB3L1, SPARC, TENT5A, MBTPS2, MESD, KDELR2) and 5 OI related genes (
PLOD2, P4HB, SEC24D, PLS3, LRP5) were included in the sequencing panel. Sequencing results from 167 out of 221 patients were previously published by our team [
2], while the other 54 were newly tested cases (Additional file
2). According to the same criteria [
2], the single nucleotide variants in
COL1A1 and
COL1A2 were classified into variants with a qualitative impact (missense) and those with a quantitative (variants leading to stop codons, splicing, or frameshift) impact.
Statistics
Depending on evidence of onset ages, the scoliosis group was divided into early-onset (EOS), late-onset (LOS), and ‘unknown’, based on radiographic evidence and a recommended consensus of demarcation point at the age of 10 [
20]. The EOS group had radiographic evidence of scoliosis before the age of 10. The LOS group had radiographic evidence of no scoliosis up to 10 or above, and that of scoliosis afterwards. The “unknown group” refers to the scoliosis patients, information of whose spinal condition before the age of 10 years old is missing. This group may thus contain both early and late onset patients.
To estimate the progression rates, we first excluded the 40 postoperative data-points in the 25 patients with surgical intervention on the spines, and obtained 606 preoperative Cobb angles of major curves in 290 patients. The progression rate was calculated as (angle difference)/(age difference in years) between successive data-points of the same patient. For the first data-point in each patient, this was calculated as (first angle)/(first age), which effectively assumed a constant progression rate since birth. The mid-point of the two ages associated with the two adjacent radiographs was used as the age corresponding with the progression rate estimate.
The statistical analyses were conducted on the R platform (version 4.0.0). For dichotomous quantitative variables, Student's t-test was used for univariate analyses. For those with multiple groups, one-way ANOVA was used. For categorical variables, Pearson’s Chi-squared test was used. For multivariate analyses with dichotomous dependent variables, logistic regression was used. The progression rates with respect to age were fitted using a local polynomial model (LOESS). In all cases, p-values were reported rounding to three digits of significance, and P < 0.05 was considered statistically significant.
Discussion
Scoliosis is one of the most prevalent conditions among osteogenesis imperfecta patients and is well-known to be progressive with age [
11,
12]. It has a huge impact on the quality of life among OI patients, yet both its disease causes and courses, which are vastly different from other more common forms of scoliosis, remain poorly understood. Previous studies of OI scoliosis either had small sample sizes [
16] or covered incomplete genotypes [
12]. In this study, we reported a retrospective study of OI scoliosis outcomes and progression based on a large cohort of 290 OI patients, of whom up to 76% had confirmed genetic information.
We stratified the cohort by four scoliosis outcome grades, including non-scoliotic, mild, moderate and severe, as measured by their maximum (preoperative, if any) Cobb angles of the major curves. We then performed univariate and multivariate analyses between the outcome and a set of genetic and non-genetic factors. We found that patients with COL1A1 and COL1A2 genotypes were strongly biased towards having mild or no scoliosis at all, whereas patients with pathogenic variants on IFITM5, WNT1 and other recessive genes did not display such a pattern. Due to the relatively small number of cases in IFITM5 and recessive genes, it was difficult to statistically delineate their effects on the outcomes, although their fractions of moderate or severe cases were comparable.
Within the two collagen genes,
COL1A2 was less damaging than
COL1A1 in progressing into advanced stages of scoliosis. The mutation types, in terms of qualitative or quantitative changes, had a weak influence on the outcomes. Neither within each collagen gene nor the two genes combined did the mutation types have a significant association with the severity grades. Among the non-genetic factors, we found that skeletal maturity, lower-limb deformities, and drug history were all individually associated with severity outcomes (Table
4), although when taken together into a multivariate logistic regression model, many of them (including drug history and lower-limb deformities) had weak or no associations (Table
5). A likely explanation is that OI scoliosis is highly age-dependent, thus the contributions of many age-dependent factors, such as skeletal maturity and LLDs, were largely absorbed by the age variable itself. Drug history may reflect patient age too, as older patients were more likely to either take pamidronate or even did not take drugs at all.
We included all patients, even those considered non-scoliotic at the cutoff date, for estimating the progression rates, which were estimated by dividing the angle difference by the age difference between successive datapoints of the same individuals. In fact, 40 of the 85 patients without scoliosis had multiple follow-ups, all of which had Cobb angles < 10°. All datapoints from these patients were included, to ensure that the progression rate estimation was unbiased and accurate. We estimated an overall progression rate of 2.7°/year, which was highest among adolescents and young adult age-groups (10–20 years), and was lower in COL1A2 than in COL1A1.
Overall, we noted both the outcome grades and progression rates were more severe in our cohort than in the literature. At 70.7%, the prevalence of scoliosis was considerably higher in our cohort than previously reported [
11,
12,
16]. The overall progression rate of 2.7°/year was in close range to but also slightly higher than previously reported estimates of 2.3–2.6°/year [
11,
12]. Both of these could be due to the inclusion of non-
COL1A1/2 patients in our study and socioeconomic reasons. Unlike in the West where 85–90% of the OI patients seeking medical treatments were affected by
COL1A1/2 mutations [
28,
29], our cohort and others in China consistently included 2/3–3/4 of such patients only [
30]. Subtype I, the clinically mild form of OI, only made up 13.4% of our cohort, whereas they usually accounted for ~ 40% in western cohorts [
11,
12,
16,
31]. At an average age of 12.0 years corresponding to the maximum Cobb angle, our cohort is also older than the study that reports 54% prevalence at a mean age of 7 years [
11].
We also studied the onset age of scoliosis in OI. Based on radiographic evidence, we stratified the scoliosis group into early-onset (EOS), late-onset (LOS) and ‘unknown’. The ‘unknown’ group represented over half of all scoliotic OI (108 out of 205, Table
2), and was thus labelled because of the missing information regarding their spinal condition before 10. We found that there are > 5 times more EOS than LOS in our cohort. Overall, the ‘unknown’ group, the exact onset age among whom cannot be confirmed, also behaved quite similarly to the EOS (Fig.
1C). In fact, there were more patients who were diagnosed with scoliosis before the age of 5 (n = 22) than after 10 (n = 15) (Table
2). We postulated that per the current consensus of cutoff age of 10 for EOS/LOS, majority of the scoliosis OI patients seeking treatment in our hospital may fall into the EOS category. Our results suggest it is important to monitor the spinal health of these patients, even though most of their medical interventions currently focus on the limbs and bone densities.
We are aware that multiple other factors may limit the accuracy in our study. Manual reading of Cobb angles and nonstandard radiographic positioning may add noise the data. We found that 52 of the 606 preoperative Cobb data-points were smaller than their immediate previous follow-ups, with an average reduction of 5.6 ± 3.7 degrees among them. Upon reexamining the radiographs, we confirmed that all of these were real, and that all but one were posture-induced. Since 48% of our cohort had LLD and other lower-limb conditions were common, standard upright radiographic postures were difficult to attain for many patients, which in turn caused considerable difficulties in accurately reading Cobb angles, even for experienced physicians. There was inevitably a certain amount of data noise attributable to such cause. The only other case involved a girl who experienced a Cobb angle drop of 18° over the course of 2 years. Re-examining the records showed that the girl had been wearing bracing for 2 years, after which the Cobb angle appeared reduced and stabilized (Additional file
1: Figure S2). Bracing has proven positive effects on other common forms of scoliosis during adolescence [
32], but its use in OI has been disputed. Early studies suggested bracing was not effective in OI scoliosis [
33,
34], and as such it was not used often in our cohort (< 10 patients) and other recent studies [
11]. As our case showed and as noted in [
11], with the use of modern anti-osteoporotic agents such as bisphosphonates, a second look into the effects of orthosis in OI scoliosis is needed in future studies.
Although we tried to make our cohort as representative as possible by including all consecutive cases, the non-
COL1A1/2 patients still only represented a minority (~ 1/3). Since eight genes were involved among these cases, the number of cases per genotype was rather small. This distribution bias may cause difficulty in estimating genotype-specific effects on OI scoliosis. We addressed this by a two-step approach, whereby the two collagen genes were first considered as a single group, before a second analysis on the 145 patients affected by these two genes only was conducted, where
COL1A1 and
COL1A2 were now treated as separate groups. Sillence classification was often used as an independent variable to explain the scoliosis outcomes [
11,
12], although it is well-known that scoliosis itself was part of the criteria for grading the Sillence subtypes [
7]. To avoid circularity, we did not present the results of analyses using it as a covariate.
Lastly, it is noteworthy that the cases in the current study only represented OI patients seeking treatment at our hospital, as a result of which some milder cases not needing medical treatment were not screened. Our results thus may appear more severe than the actual situation among the broader OI community.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.