The skeletal system is the third most common metastatic site of most cancers (Sutcliffe et al., 2013). Postmortem examination showed that about 70% of cancer patients had spinal metastases (Bartels et al., 2008). Intractable pain, neurologic deficits, and paralysis are common symptoms of these patients, fundamentally impacting their survival and quality of life (Tang et al., 2016). Survival time of the patients with bone metastases is a key consideration for the decision-making of subsequent treatments. However, there is a lack of robust clinical tools or biomarkers to accurately predict the prognosis of bone metastatic patients.

Several scoring systems including Tomita, Sioutos, and Van der Linden scores have been developed (Tomita et al., 2001; Sioutos et al., 1995; van der Linden et al., 2005), and the revised Tokuhashi score is the most commonly used in survival prediction of cancer patients (Tokuhashi et al., 1990; Tokuhashi et al., 2005). However, previous studies have demonstrated that the revised Tokuhashi score is not accurate enough in prognostic prediction of patients with spinal metastasis, and therefore the survival of patients with low a Tokuhashi score (≤8) might be underestimated (Tan et al., 2016; Yang et al., 2019). Precise selection of these patients could improve therapeutic planning and ameliorate the clinical outcomes.

The implementation of molecular diagnostics involves scalability, sensitivity, and specificity. In patients with spinal metastasis, whole-genome sequencing analysis of cell-free DNA has emerged to be a promising strategy for reflecting the burden and genetic alterations of tumors (Ellinger et al., 2011). The profiling of genetic alterations in circulating cell-free DNA from patient plasma has potential clinical applications including early disease detection, real-time prediction of treatment response, and prognostication (Tran et al., 2021). In various cancer types, cfDNA unveils abundant information including the metastatic status, microsatellite instability, somatic copy number alterations, and the structural rearrangements (Vanderstichele et al., 2017; Mayrhofer et al., 2018). Copy number variation (CNV) is the result of ongoing changes in the number or structure of chromosomes, which was a hallmark in various cancers (Vanderstichele et al., 2017; Gronroos and López-García, 2018). Initially, maternal neoplasia was incidentally detected by identifying chromosomal and/or subchromosomal abnormalities during noninvasive prenatal testing (Bianchi et al., 2015; Dharajiya et al., 2018). A later study on the large population validated a novel cancer detection pipeline algorithm to enhance the detection of occult maternal malignancies through identifying multiple chromosomal aneuploidies (Ji et al., 2019). In addition to noninvasive prenatal testing, other noninvasive examinations focusing on chromosomal instability analysis suggested high CNV detected in cfDNA could serve as surrogates for predicting and monitoring malignancies (Oellerich et al., 2017).

In this study, low-coverage whole-genome sequencing of cfDNA was conducted to examine blood plasma from patients with spinal metastasis. An analysis pipeline was developed and validated to evaluate the CNV status in cfDNA, in an attempt to determine whether the CIN score, which was defined as a prognostic factor in patients with metastatic cancer like breast cancer (Mo et al., 2020), could also predict the survival of pan-cancer patients with spinal metastasis and especially identify the patients with actual good prognosis but ranked with a low Tokuhashi score.

Spinal metastasis is a common occurrence in multiple advanced cancers (Bartels et al., 2008). Pretreatment survival prediction may help clinicians in the decision-making process for subsequent treatments. Several survival predictive models have been developed over the past decades, but their accuracy is unsatisfactory (Tan et al., 2016). On the other hand, there is a lack of real-time biomarkers with consideration of the molecular characteristics of cancer cells for survival prediction (Bauer et al., 2002). It was found in this study that CNVs from cfDNA could act as a potential biomarker for prognosis prediction of patients with spinal metastasis and help discriminate long-survival patients from poor-prognosis patients assessed by traditional prediction models in the perspective of molecular characteristics of malignancies.

Copy number variation is a hallmark in various cancers, including metastatic cancers, and deemed a prognostic predictor associated with an increased risk of recurrence or death (Gronroos and López-García, 2018; Tan et al., 2019). In this study, cfDNA CNV was first proved to be a potential biomarker for predicting the survival of patients with spinal metastasis. Vanderstichele et al. (Vanderstichele et al., 2017) reported that evaluating the cfDNA with CNV could help distinguish the patients with an adnexal mass caused by an invasive carcinoma from patients presenting with a benign adnexal mass. In addition, more genetic and epigenetic alterations could also be detected by cfDNA sequencing (Ellinger et al., 2011). Mehra et al. (2018) demonstrated that the cfDNA fragment concentration was an independent prognostic variable for radiological progression-free survival and OS of patients with metastatic prostate cancer. Stover et al. (2018) reported that tumor fraction was associated with worse survival in metastatic triple-negative breast cancer patients, and specific somatic copy number alterations were enriched in these patients. Spinal metastases originating from multiple types of cancer are featured by heterogeneous molecular characters, which impose a significant challenge on clinicians in precise diagnosis, evaluation, and treatment (Jansson and Bauer, 2006; Kawahara et al., 2009). As aforementioned, CNV could present the common properties of metastatic malignancies. Our results demonstrated that CNV in cfDNA was significantly associated with the survival of patients with spinal metastasis, which provides a universal and convenient tool of pretreatment evaluation for different cancers as an integrated entity. This interpretation of CNV could be applicable to the surveillance of progression of malignancies in other scenarios.

Another intriguing finding of our study is that CNV successfully discriminated the good-prognosis patients with low Tokuhashi score from poor-prognosis patients. According to the revised Tokuhashi score, the predicted survival of patients with a total score ≤8 was less than 6 months (Tokuhashi et al., 2005). However, it was found that Tokuhashi score might not be qualified to establish a precise prediction of survival, especially in predicting short-term survival (Ahmed et al., 2018). Some patients ranked with a low Tokuhashi score may possess a relatively longer survival than expected. Ahmed et al. (2018) reported that the revised Tokuhashi score was not accurate in 90-days survival prediction after surgery, with the AUC being 0.67. The data of Tan et al. (2016) suggested that only 41.7% patients had a Tokuhashi-predicted survival that was correlated with actual survival. Judged by Tokuhashi score, only 6.1% patients were predicted to survive more than 6 months, while 44.4% of patients in the cohort attained survival longer than half a year. Moreover, the mean OS of patients with spinal metastases have been observed to be improved and longer than 1 year in the recent literature (Tang et al., 2015; Yang et al., 2019), which is in congruence with our data. As CNV level can reflect the disease progression through the actual activity of the tumor individually, it proved capable of predicting the prognosis of patients with highly malignant spinal metastases in the current study, including lung cancer. Identification of the patients who have long-term survival in the low Tokuhashi score group can contribute to a more positive alteration of treatment, thus improving their actual prognosis. Pelegrini de Almeida et al. (2018) retrospectively analyzed 117 patients and found that the patients with low Tokuhashi score for whom surgical treatment was not recommended traditionally had better quality of life and longer survival after adequate surgery than the results inferred by the Tokuhashi system. It was found in our study that patients with poor Tokuhashi score but low CNV had relatively good prognosis, with the median OS of 433 days. Accurate survival prediction of this part of patients through cfDNA test might optimize the selection of their treatment procedures.

Unlike the traditional predictive models for metastatic malignancies such as the Tokuhashi score, which may be influenced by subjective confounders, CNV is calculated based on the plasma cfDNA, which represents a more objective set of criteria. Almost all previous models focus on clinical characteristics of the patients with spinal metastases, while CNV reflects the tumor status of progression in circulation (Batista et al., 2016; Vanderstichele et al., 2017). Although other blood tests including certain tumor biomarkers (CEA, AFP, etc.), inflammation-based biomarkers (platelet–lymphocyte ratio, neutrophil–lymphocyte ratio, etc.), and hemostasis biomarkers (D-dimer, fibrinogen, etc.) were also reported to be associated with survival of cancer patients, they are either specific only for a single tumor type or lack sensitivity (Ay et al., 2012; Li et al., 2018; Yang et al., 2019). Examining CNV in cfDNA for prognosis prediction provides more information on the innate and dynamic character of metastatic malignancies, which are often ignored in most traditional models. Furthermore, accurate survival prediction through CNV may help make a more precise treatment decision. It has been deemed as the principle that the choice of treatment should be judged by the prognosis of spinal metastatic patients (Tokuhashi et al., 1990; Tomita et al., 2001). For good-prognosis patients, radical excision with wide or marginal margin is suggested for long-term local control, while for patients with poor prognostic prediction, intralesional excision or palliative treatment can reduce unnecessary iatrogenic injury (Tomita et al., 2001).

The main limitation of this study is that all blood samples were obtained from the cohort in one single institution. The patients recruited in our study are relatively homogeneous, and were all diagnosed with spinal metastases, implying that our conclusion should be interpreted cautiously in patients with metastatic sites involving other systems. Besides, we noticed different features of CIN and different clinical outcomes between tumors from various origins, but the score was worse when estimating each carcinoma (Supplementary Figure S1) than that of pan-cancer as a whole (Supplementary Figure S1). Thereby, an increased number of patients and further investigation taking into account the differences between malignancies are required to improve the performance of the algorithm. Second, as the cohorts of the discovery group and validation group were not evaluated synchronously, relevant biases may not be avoidable. The constitution of the malignant types is somewhat different between the two groups; for instance, fewer patients in the validation group were diagnosed with cancers of unknown primary sites than those in the discovery group (p < 0.001). Third, clinical advances in imaging and histopathological examinations with specific immunohistochemistry in more recent years may also affect tumor diagnosis and increase the sensitivity of primary tumor detection (Collado Martín et al., 2014; Varadhachary and Raber, 2014). Finally, more female patients were included in the validation group than those in the discovery group (p = 0.02).

In summary, we proposed a framework for the minimally invasive examination of genetic architecture of spinal metastatic cancers. The interpretation of its result was proved to be correlated with the clinical outcome of patients. A high cfDNA CIN score suggests a dismal survival of patients with spinal metastasis. cfDNA testing for CNV could be applied into clinical practice as a new indicator for the prognosis of spinal metastasis. Accurate selection of patients with either predicted long or short survival might improve the treatment planning and ameliorate the clinical outcome.

A CIN score as a new quantified survival indicator is developed for survival prediction of patients with spine metastasis.

The skeletal system is the third most common metastatic site of most cancers. About 19.7% of the terminal-stage cancer patients with metastasis diagnosed during 2015–2018 survived more than 2 years, and over half of them were still alive until the last follow-up in our center. However, the widely used revised Tokuhashi score lacks accuracy in prognostic prediction. About 50% of the patients with a Tokuhashi score under 8 survived for more than 1 year but were tortured by both physical suffering and mental desperation. In this study, we present a CIN score as a new quantified indicator for the survival prediction of such patients. The CIN score is a more objective set of criteria that can reflect both molecular characteristics of all cancer subclones and tumor burden instead of traditional symptoms and clinically oriented scores. Therefore, it can ensure the optimal therapeutic efficiency by helping make a more precise clinical decision.