Patients

The database of a tertiary hospital center with a comprehensive cancer program was searched for patients diagnosed with lung metastases (ICD-9-CM code 197.0) over the years 2010–2020. The records of these patients were reviewed for the following parameters: age, gender, type of malignancy, date of cancer diagnosis, date of metastases diagnosis, and fate of the patients. The study was approved by the IRB of the Hadassah Medical Organization (# 0650-21-HMO) All data were fully anonymized before analysis and the IRB waived the requirement for informed consent.

Patients included in the study had:

1. Diagnosis of one of the following primary tumors: colorectal, breast, kidney, thyroid, bladder, prostate, thyroid, sarcoma, or melanoma.
2. 2. Two or more lung metastases defined as discrete, round or ellipsoid lesions inside the lung parenchyma, measuring ≥3 mm in their largest diameter.

Exclusion criteria:

1. Patients with history of primary lung cancer.
2. Patients with more than one primary cancer.

The axial chest CT scans of 503 patients, obtained using voltage setting of 120KVp, amperage of 260–330 mA and slice thickness reconstruction of 3 mm or less, were reviewed. The patient’s last CT was selected for interpretation. Nonetheless, earlier surveys were examined in 26 patients (5.2%), whenever the last CT was uninterpretable due to confounding variables such as large pleural effusion, lung collapse, coalescence of multiple metastases that precluded their individual measurement or whenever curative therapy was successful (e.g., radioactive Iodine in metastatic carcinomas of the thyroid). For each study, metastasis largest diameter was determined manually, by placing electronic calipers at the margins of the tumor. The number and the largest diameter of all the metastases were recorded.

Statistics

For each patient, lung metastases that appeared in clusters of similar diameters (i.e., with a diameter deviation ≤ 1 mm) were counted and their sum number was marked as ∑c. The sum number of metastases that cannot be classified (isolated) into a similar size category was marked as ∑i. The linear/parallel ratio (LPR) was calculated for each patient using the formula (∑c-∑i)/(∑c+∑i) with the aid of a Python written computer code (S1 File). LPR ratio of 1 suggests pure linear progression and -1 pure parallel.

Pancreas

Pancreatic tumors showed an almost pure linear progression pattern with a median LPR of 1. This is supported by genome sequencing of primary pancreatic tumors and their metastatic deposits showing that the clonal population of metastases is represented in the primary tumor [14]. This implies that metastases are a late event in pancreatic cancer evolution, and that early diagnosis and local treatment are potentially curative. Unfortunately, thus far there is no effective mean for early detection of pancreatic cancer.

Prostate

Here too, a median LPR of 1 suggests that metastatic spread in prostate cancer is almost exclusively by the linear mode. This is in agreement with molecular studies. Primary prostate cancer is a multifocal disease harboring multiple genomically distinct foci with non-overlapping mutation profiles, suggesting separate and independent evolutionary trajectories [15, 16]. Prostate cancer metastases, on the other hand, show mono or oligoclonality [16–19]. The linear model in prostate cancer is also supported by the clinical findings of the STAMPEDE trial [20]. This study showed that prostate radiotherapy improves overall survival of men with low metastatic burden (presumably, after only one or a few waves of metastases were deployed) by preventing further waves. In patients with high metastatic burden (after multiple waves of metastases), this treatment is less efficient.

Renal cell Carcinoma (RCC)

The median LPR of kidney cancer was 0.91 pointing toward dominance of the linear mode of progression in this disease, but with existence of at least an occasional component of parallel dissemination mode as well. This is supported by genomic analyses of 575 primary and 335 metastatic lesions in 100 RCC patients by Turajlic et al. These authors showed two types of metastatic progression: rapid progression to multiple tumor sites by primary tumors of monoclonal structure (linear) and attenuated progression by heterogenous cells (parallel) [21].

The CARMENA clinical trial randomized intermediate and poor-risk metastatic RCC patients to Sunitinib only or to Sunitinib plus cytoreductive nephrectomy. Sunitinib alone was non-inferior to cytoreductive nephrectomy followed by sunitinib [22]. However, significant overall survival benefit (HR 0.34, 95% CI, 0.22–0.54) was gained in patients who underwent secondary nephrectomy, stressing the importance of the primary tumor as a continuous source of further waves of metastases as demonstrated by the linear model.

Bladder

The median LPR of bladder cancer was 0.50, one of the lowest in the study, suggesting a predominant role of parallel metastasis progression. This is in agreement with studies of genomic classification comparing primary tumors and their metastases. Faltas et al, performed whole exome sequencing and clonality analysis of 16 matched sets of primary and metastatic tumors. They found that metastatic spread was multiple, parallel and occurred early in the natural history of the disease in all cases [23]. The origin of this phenomenon is probably the high mutation burden of this tumor leading to a high interpatient, intratumoral and intertumoral heterogeneity [24]. Additionally, neoadjuvant chemotherapy before radical cystectomy improves survival and is the standard of care in patients with non-metastatic muscle-invasive bladder cancer, further supporting the concept of early, subclinical spread of invasive bladder cancer cells, requiring more than local excision [25].

Colorectal cancer (CRC)

The current study showed mixed parallel and linear progression patterns of CRC (median LPR 0.60). This is in concordance with genomic analyses of primary colon cancer and its metastases which likewise show conflicting results suggesting that both mechanisms of metastatic spread are active in CRC. Wei et al, using whole-exome sequencing of 28 samples of matched primary and metastatic tumors from four CRC patients showed that all metastases inherited multiple genetically distinct subclones [26]. Similarly, Vermaat et al, using deep sequencing of DNA isolated from primary and subsequent hepatic metastasis of 21 patients, showed substantial genetic differences between the primary tumor and its metastases, with an average gain of 83 potentially function-impairing variants and 70 losses [27].

Breast

Our study suggests that dissemination of breast cancer is complex and likely includes both linear and parallel progression (LPR 0.58). This is in line with phylogenetic analysis of primary and metastatic breast cancer and with the findings of high level of DTCs with high heterogeneity in the bone marrow [28–30]. It is also supported by the clinical recommendation for neoadjuvant chemotherapy [31]. A parallel model may explain the failure of locoregional treatment to affect overall survival of some women with breast cancer [32]. Thus, distinction of pattern may be of particular importance for the treatment paradigms of this cancer.

Melanoma

This tumor shows a mixed linear and parallel progression pattern (LPR 0.65). Surgical treatment in early stages of melanoma is associated with excellent prognosis (98.4% 5-year overall survival for stages 0, 1, and 2), suggesting that early parallel spread does not occur in most patients. Yet, in metastatic patients it has been shown that lineage diversification is pervasive, supporting the parallel mode of metastatic spread [33]. Melanoma can also switch between proliferative and invasive states according to environmental conditions further contributing to a disorganized distribution of metastases’ diameters in this tumor [34].

Sarcoma

The LPR of sarcomas was the lowest among the tumors studied (median 0.43), suggesting mixed linear and parallel dissemination. This is in concordance with whole exome and genome analyses of 86 tumor regions in 10 patients with metastatic osteosarcoma. Metastases showed significantly higher mutational burden and genomic instability compared to the primary tumors. Pulmonary spread was linear in 6/10 patients and parallel in 4/10 [35]. It is suggested that neoadjuvant chemotherapy for treating micrometastases be considered before surgical treatment of sarcomas.

Thyroid

Clinicians have long been aware of the striking number and resemblance of lung metastases in thyroid cancer that are often very small and termed micronodular metastases (Fig 2). This is reflected by a high LPR (median of 1), indicating an almost pure linear progression pattern. Unfortunately, genomic information in this disease is limited. A single study in whole-exome sequencing showed a median concordance of only 38% in somatic mutations between the metastases and the primary tumor but only 3 patients with lung metastases were included in this series [36].

The LPR model is probably an oversimplification of the metastatic process (as any model in biology), yet it can potentially explain, predict and refine phenomena in oncology. A potential example is the oligometastatic state (OS), usually defined as a state of five metastases or less [37, 38]. Cases of oligometastasis can potentially be managed by metastasis-directed therapy without systemic treatment. Yet, patients with parallel spreading tumors, even with less than five detectable metastases are obviously not suitable for local treatment only. Adding a high LPR to the definition of OS will decrease the number of patients with OS but will improve their outcome and will direct patients with non-OS to systemic therapy. As stated previously, formal calculation of the LPR is not mandatory, and the naked eye can easily categorize similar and dissimilar metastases (Fig 2).

Limitations of the study

Although potentially helpful in the many ways, the LPR model has several limitations worth mentioning:

1. Metastases to the same organ originating in a single clone may grow in different rates due to proximity to nutrients or to an anatomical structure, thus reaching different diameters and falsely decreasing the LPR.
2. In a parallel spreading disease, a later, nevertheless faster growing clone could reach the same size as an earlier but slower growing one, thus reaching the same diameter and falsely increasing the LPR.
3. For the sake of simplicity, metastases diameters were measured from the axial CT slices only; however, this may not necessarily be their largest diameter thus distorting the measurements.
4. The conclusions of this work apply to metastatic implantation in the lung microenvironment. It is not certain that they can be extrapolated to other organs.
5. The analysis is population derived and based on a snapshot and not on sequential images in individual patients that could reveal a different course.
6. Histologic confirmation of CT findings was not performed in most of the cases. However, 81% of the patients in the series died during follow-up and death was disease-specific in 98.7% of them, implying that the lesions seen on chest CT were indeed metastases in most cases.
7. The LPR was calculated using metastases longest axis on the axial views. This is not necessarily the best morphological predictor of the LPR. Other potential parameters such as metastases short axis, multiplication of the axes or perhaps metastases volume may deliver better predictions. We hope that future research will provide this information.

In summary, we showed a high LPR for metastases originating in the thyroid, kidney, pancreas, and prostate. This point toward a linear route of metastatic spread. The LPR of melanoma, colorectal, breast, bladder, and sarcoma points toward mixed linear and parallel routs with increasing relevance of the parallel route in this order. These findings are in line with most of the genomic work performed and with clinical observations. The LPR was calculated here with aid of a computer code, but can easily be observed by the naked eye, which is sensitive to this type of information. Thus, with further confirmation, we envision that the LPR can be applied to individual patients and assist in oncological decision-making. Future directions of studying the LPR model may include verifications of the findings in other databases and other organs (brain and liver metastases), genomic analyses of several metastases removed from the same patients to study if similar size equals similar genomic traits and comparing the growth of individual metastases in sequential CT studies.