Intrathecal trastuzumab versus alternate routes of delivery for HER2-targeted therapies in patients with HER2+ breast cancer leptomeningeal metastases

Background Patients with HER2+ breast cancer (BC) frequently develop leptomeningeal metastases (LM). While HER2-targeted therapies have demonstrated efficacy in the neoadjuvant, adjuvant, and metastatic settings, including for parenchymal brain metastases, their efficacy for patients with LM has not been studied in a randomized controlled trial. However, several single-armed prospective studies, case series and case reports have studied oral, intravenous, or intrathecally administered HER2-targeted therapy regimens for patients with HER2+ BC LM. Methods We conducted a systematic review and meta-analysis of individual patient data to evaluate the efficacy of HER2-targeted therapies in HER2+ BC LM in accordance with PRISMA guidelines. Targeted therapies evaluated were trastuzumab (intrathecal or intravenous), pertuzumab, lapatinib, neratinib, tucatinib, trastuzumab-emtansine and trastuzumab-deruxtecan. The primary endpoint was overall survival (OS), with CNS-specific progression-free survival (PFS) as a secondary endpoint. Results 7780 abstracts were screened, identifying 45 publications with 208 patients, corresponding to 275 lines of HER2-targeted therapy for BC LM which met inclusion criteria. In univariable and multivariable analyses, we observed no significant difference in OS and CNS-specific PFS between intrathecal trastuzumab compared to oral or intravenous administration of HER2-targeted therapy. Anti-HER2 monoclonal antibody-based regimens did not demonstrate superiority over HER2 tyrosine kinase inhibitors. In a cohort of 15 patients, treatment with trastuzumab-deruxtecan was associated with prolonged OS compared to other HER2-targeted therapies and compared to trastuzumab-emtansine. Conclusions The results of this meta-analysis, comprising the limited data available, suggest that intrathecal administration of HER2-targeted therapy for patients with HER2+ BC LM confers no additional benefit over oral and/or IV treatment regimens. Although the number of patients receiving trastuzumab deruxtecan in this cohort is small, this novel agent offers promise for this patient population and requires further investigation in prospective studies.


Introduction
Leptomeningeal metastasis (LM), also known as neoplastic meningitis or leptomeningeal carcinomatosis, is a debilitating condition associated with advanced breast cancer (BC) [1]. LM is defined by cancer cells reaching and proliferating in the subarachnoid space that surrounds the brain and spinal cord [1]. The development of LM portends a dismal prognosis for BC patients, with median overall survival (mOS) measured in weeks to months [2].
For patients with HER2+ BC LM, no randomized controlled trials have been performed comparing HER2-targeted regimens. However, seven single-armed prospective studies assessing HER2-targeted therapies for BC LM have been published to date [5][6][7][8][9][10][11]. Three of these trials studied the safety and activity of intrathecal (IT) trastuzumab [5][6][7]. This has led to uptake of this treatment approach in many centers globally despite its modest but clear morbidity [7]. It remains uncertain whether IT administration of trastuzumab confers any tangible benefit for patients with HER2+ BC LM compared to other approaches with respect to meaningful clinical endpoints, such as quality-of-life, progression-free survival (PFS), and overall survival (OS).
To evaluate the efficacy of HER2-targeted therapy in the management of BC LM, we performed a systematic review and meta-analysis of all published data on clinical outcomes in patients with HER2+ BC LM treated with HER2-targeted therapies. This has allowed us to make the first comparisons between IT versus intravenous (IV) or oral HER2targeted therapies for BC LM and present the first evidence supporting the efficacy of T-DXd compared to alternative strategies for this patient population.

Methods
Search Strategy: A literature search was conducted of studies published from January 1964 to December 2021 in the following databases: Medline ALL (Medline and Medline Epub Ahead of print and In-Process & Other Non-Indexed Citations), Embase, Cochrane Central Register of Controlled Trials, Scopus, and Web of Science Core Collection. The detailed search strategy is presented in Appendix 1. Published conference abstracts were included. Additional publications and/or data identified by the authors outside of the search were added to the systematic review when applicable. The study protocol was prospectively uploaded to PROSPERO (ID: CRD42021292539) and followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines (PRISMA) [12,13].
All abstracts were screened by two independent reviewers (reviewers included authors AML, SMM and MD) using the Rayyan software (www. rayyan.ai). Conflicts were resolved with internal discussion between the three reviewers (AML, SMM, MD). For any publications for which a consensus could not be reached (N = 5 abstracts), a fourth reviewer (NB) made the determination to include or exclude. Of the 32 articles extracted by the authors of this manuscript without obtaining data from authors of the source manuscript, 50% (16) articles were independently extracted by two reviewers (AML and MD), revealing 100% concordance between both reviewers. The remaining 16 articles were extracted by a single reviewer (AML).
After data was extracted from all included publications, missing data was identified and requested from the original authors of each publication with two separate email prompts >7 days apart. Clinical information from 93 patients, corresponding to 117 lines of therapy, were extracted by the authors of this manuscript from the source publications, while data for 115 patients, corresponding to 158 lines of therapy, were obtained through communication with the authors of the corresponding publication (Appendix 2). We unsuccessfully attempted to obtain data from an additional 16 retrospective and 7 prospective studies, corresponding to 343 patients (Appendix 3). Together, we captured approximately 38% of HER2+ BC LM patients treated with HER2targeted therapies described in the literature. However, this is likely an underestimate of the true percentage captured because the number of patients described in many of the studies we unsuccessfully attempted to obtain data from describe general patient populations that would require further refinement based on our inclusion and exclusion criteria.
Inclusion and exclusion criteria: Inclusion criteria were as follows: adult HER2+ BC patients (aged 18 years or older), defined by 3+ immunohistochemistry (IHC) staining or 2+ IHC with fluorescence in situ hybridization (FISH) demonstrating HER2 amplification, with a diagnosis of LM defined on magnetic resonance imaging (MRI) or with positive cerebrospinal fluid (CSF) cytology and receiving a HER2targeted therapy for the treatment of LM. HER2-targeted therapies evaluated included trastuzumab (IT or IV), pertuzumab, lapatinib, tucatinib, neratinib, T-DM1 and T-DXd (Supplemental Table S1). Hormone receptor positivity was defined by the authors of each individual study incorporated in the meta-analysis. There was one male patient included in our study. Radiotherapy (RT) employed for the treatment of LM included stereotactic radiosurgery (SRS), whole-brain RT (WBRT) and spinal RT.
Primary and secondary outcomes: The primary outcome was OS, which was calculated from the start of HER2-targeted therapy for LM. The secondary outcome was CNS-specific PFS, which was calculated based on central nervous system (CNS)-specific progression, or death. Progression was defined by the primary paper's author's assessment via MRI or CSF analysis, or death of the patient. Statistical analyses in Tables 2 and 3 were performed with available individual patient data.
Quality (risk of bias) assessment: To assess the methodological quality of individual studies included in the meta-analysis, we used a previously described tool that is adapted for evaluation of case reports and case series [14]. The tool includes five items that are derived from the Newcastle-Ottawa scale. These five items examine the selection and representativeness of cases and the ascertainment of outcomes and exposure, with each item scored one point if the information was specifically reported. The study was deemed to be of good quality (i.e. low risk of bias) when all five criteria were met (score of 5), of moderate  quality when four criteria were met (score of 4), and of poor quality/high risk of bias when 3 or less criteria were met (score of 3 or less) [15].
Statistical analyses: We performed one-stage meta-analyses of pooled individual patient data from all included studies. Patient characteristics were compared between those who received intrathecal trastuzumab and those who did not, with Fisher's Exact test and Pearson's X 2 . The hazard ratio (HR) was used as the parameter of interest for OS and CNS-specific PFS. Cox proportional hazard models were used to determine the HR between groups of interest and its associated 95% confidence interval (CI). A multi-level mixed-effects Cox proportional hazards model, incorporating individual study as a random effect, was used to estimate the HR, its associated 95% CI, and P-value.
Multivariable Cox proportional hazards regression models were used to estimate adjusted OS and CNS-specific PFS (aOS and aPFS), also with a multi-level mixed-effects Cox proportional hazards regression model that incorporated individual study as a random effect. All variables with P < 0.05 in univariable analysis were incorporated into the initial multivariable model. We performed backward stepwise selection to remove insignificant variables. The final model included all variables with P < 0.05. For aOS, the initial multivariable model included geographical location (Europe), patient sample size in the study, age, hormone receptor status, lines of therapy in the metastatic setting, Eastern Cooperative Oncology Group (ECOG) status, treatment with neratinib and treatment with T-DXd versus T-DM1. For CNS-specific aPFS, the initial model included geographical location (Asia), patient sample size in the study, lines of therapy in the metastatic setting, status of prior HER2-targeted therapy, ECOG status, diagnosis with MRI, treatment with T-DXd, treatment with lapatinib, and treatment with T-DXd versus T-DM1. For both CNS-specific aPFS and aOS, this left only ECOG status as the only statistically significant variable in the multivariable model. We subsequently performed a sensitivity analysis by adding our variable of interest (IT versus IV/oral administration of HER2-targeted therapies) to the model, to obtain our final multivariable model. We tested the proportional hazards assumption by plotting the Schoenfeld residuals for each univariable and multivariable analysis, and they appeared random. Survival curves were visualized and evaluated with the Kaplan-Meier method and the log-rank test. Statistical analyses were performed with STATA v17 (StataCorp LLC, College Station, Texas, USA).
Correlation analyses between CNS-specific PFS and OS were performed with linear regression and Pearson's X 2 . When performance status was presented as Karnofsky Performance Status score, it was converted to ECOG using the previously described conversion scale [16].
Patient data: For the patients included in this study that were not previously included in other published reports, patients provided written consent for their medical records to be searched and included in this study in an anonymized fashion as case reports, in concordance with the Declaration of Helsinki.

Characteristics of included studies and patients
We identified 7780 potentially eligible articles in our search. After screening these articles, removal of ineligible articles and addition of studies from authors' files, a total of 45 publications were included in our review (Appendix 2). This consisted of a total of 208 patients with HER2+ BC LM (Table 1) who received a total of 275 patient-lines of therapy for the treatment of LM (Fig. 1). A risk of bias assessment was also performed for all studies included in the meta-analysis on a 5-point scale (Supplemental Fig. S1).
Of the 275 patient-lines of therapy in our cohort, 92 received IT trastuzumab and 183 received regimens that included oral or IV HER2targeted therapy (Table 1). When comparing clinical characteristics of the IT and no-IT cohorts, we observe that the no-IT cohort is enriched in patients of older age (P < 0.05), patients who were hormone receptor positive (P < 0.05), patients who were more likely to have concurrent extracranial metastases (P < 0.05), patients who were less likely to receive concurrent chemotherapy alongside HER2-targeted therapy (P < 0.001), and patients who were more likely to receive RT (P < 0.05). Patients in the IT cohort were more likely to have been reported in prospective studies (P < 0.001) and to be patients who had received fewer lines of therapy in the metastatic setting (P < 0.05). a T-DM1 and T-DXd were administered intravenously in the context of the patient receiving intrathecal trastuzumab simultaneously.

Characteristics associated with OS and CNS-specific PFS
In the entire population, mOS and median CNS-specific PFS (mPFS) in the cohort was 14.3 and 6.0 months, respectively (Tables 2 and 3). In the patients where CNS-specific PFS was available from the source publication, we observe significant correlation between CNS-specific PFS and OS (Pearson's R 2 = 0.63, P < 0.0001; Supplemental  Fig. S2A). This significant correlation remains when analyses are restricted to patients who had documented death, or whose progression was not caused by death (Supplemental Fig. S2 B-D, Supplemental Table S2).

Comparing anti-HER2-targeted therapies for BC LM
We explored whether different categories of anti-HER2 therapeutics are associated with differential outcomes. We observe no significant difference in OS and CNS-specific PFS between regimens that included monoclonal antibody-based agents (trastuzumab, trastuzumab + pertuzumab, T-DM1, T-DXd) versus those that exclusively employed HER2 tyrosine kinase inhibitors (TKI; lapatinib, tucatinib, neratinib; Supplemental Fig. S3 A-B) or whether chemotherapy was added to anti-HER2targeted therapies (Supplemental Fig S3 C-D). Moreover, the route of chemotherapy administration (IT, IV/oral or IT and IV/oral) was not associated with significant differences in OS or CNS-specific PFS (Supplemental Fig. S3 E-F).
Next, we examined whether individual agents are associated with prolonged OS and CNS-specific PFS. Trastuzumab, pertuzumab, and T-  DM1 were not associated with differential outcomes (Supplemental Fig. S4 A-F), and lapatinib was associated with prolonged CNS-specific PFS (P = 0.024) but not OS (P = 0.094) compared to other HER2targeted therapies (Supplemental Fig. S4 G-H). Treatment with T-DXd was associated with prolonged OS (P < 0.05) and prolonged CNS-specific PFS (P < 0.01) (Fig. 3A-B). Furthermore, T-DXd demonstrated superior OS (P < 0.05) and CNS-specific PFS (P < 0.01) compared to T-DM1, another antibody drug conjugate (Fig. 3 C-D). Of the 15 patients treated with T-DXd who were included in the cohort, two are previously unpublished patients from our center. Both of these patients were treated with single agent T-DXd in the absence of surgery or RT for HER2+ BC LM and experienced profound clinical and imagebased responses to treatment in their leptomeningeal lesions (Fig. 3 E-F). Both patients exhibited impressive responses lasting 16 months, one of which remains on treatment with ongoing treatment response.

Quality assessment
The majority of the patients included in this analysis were reported in retrospective studies. These patients may be subject to greater bias than patients identified from prospective studies. However, in our cohort, we observe no difference in OS or CNS-specific PFS between patients identified from retrospective versus prospective studies (HR = 1.5, 95% CI: 0.77-2.7, P = 0.25 and HR = 0.60, 95% CI: 0.19-1.9, P = 0.40, respectively) ( Table 2, Table 3, Supplemental Fig. S5 A-B).
We next classified studies according to their risk of bias using a 5point score that was adapted from the Newcastle-Ottawa scale [14].
Studies with a risk of bias (ROB) of 3 or less were classified as high risk of bias, while studies with an ROB of 4 or 5 were considered to have a moderate to low risk of bias. When comparing patients extracted from studies of moderate to low versus high risk of bias, we observe no significant differences in OS and CNS-specific PFS (HR = 1.1, 95% CI: 0.66-1.8, P = 0.70 and HR = 0.95, 95% CI: 0.60-1.5, P = 0.82, respectively) ( Table 2, Table 3, Supplemental Fig. S5 C-D). We also observe no difference in outcomes of IT versus non-IT treated patients when only including those from studies with moderate to low risk of bias in the analysis (Supplemental Fig S5 E-F).
Furthermore, it has been previously shown that BC LM patients with spinal cord involvement experience worse prognosis compared to those who have brain-only disease [17]. For this reason, we explored whether patients with spinal cord involvement of their LM were more likely to derive benefit from IT therapy. No significant difference in OS was observed between patients with spinal versus brain-only LM (P = 0.8), while there was a non-significant trend towards prolonged CNS-specific PFS among patients with spinal cord involvement treated with IT HER2-targeted therapy (P = 0.060) (Supplemental Fig. S5 G-H).

Discussion
We initiated this study because HER2-targeted therapy is routinely used in patients with BC LM despite these patients not being included in any of the randomized controlled trials studying these agents. Therefore, high quality data on the efficacy of these HER2-directed therapies for patients with LM is lacking. By extracting data from 45 publications, corresponding to 208 patients and 275 lines of HER2-targeted therapy, we developed the largest cohort of treated HER2+ BC LM patients that has been compiled to date. Our dataset demonstrates that HER2-targeted therapies have clinical activity in the setting of BC LM, with several patients experiencing durable and prolonged treatment responses. We identified no statistically significant difference in OS or CNS-specific PFS when HER2-targeted therapy is introduced intrathecally or intravenously. Several biological reasons can explain this finding. It is possible that IV trastuzumab reaches the subarachnoid space in sufficient concentrations to effectively treat LM in patients with trastuzumab-sensitive disease [18], and that the apparent resistance of BC LM to trastuzumab is largely mediated by the fact that the overwhelming majority of these patients have already received IV trastuzumab in prior lines of therapy (Table 1). It has been previously described that IV trastuzumab reaches significantly higher concentrations in the CSF of patients with LM and/or those who received WBRT compared to non-LM/non-WBRT patients [18]. However, these studies have also demonstrated that the trastuzumab concentrations in CSF are still an order of magnitude lower than serum concentrations even for LM patients receiving WBRT. Despite this, it is plausible that the microenvironmental concentrations at the site of LM lesions harboring local blood-CSF-barrier disruption approach serum levels that are sufficient to exert activity [18,19]. Furthermore, it has been demonstrated that IT trastuzumab rapidly distributes out of the CSF and into the serum, quickly negating any LM-specific efficacy that may exist with IT administration [20].
An additional benefit of employing IV over IT trastuzumab is that it would be expected to elicit greater activity for LM patients who have concurrent systemic disease. Indeed, 81% and 60% of patients who received IV/oral only and IT regimens in our dataset had extracranial metastases at the time of treatment, respectively (Table 1).
While no prospective trials of IV trastuzumab for BC LM have been published, two single-armed trials assessing the efficacy of IT trastuzumab in patients with HER2+ BC LM have been reported in the past year [6,7]. The phase II trial of IT trastuzumab (150 mg once weekly) in 19 patients with HER2+ BC with LM demonstrated a CNS-specific mPFS of 5.9 months and a mOS of 7.9 months [6]. Another phase I/II study of IT trastuzumab (80 mg twice weekly) in 26 HER2+ BC with LM demonstrated a mOS of 10.5 months [7]. Both studies did not describe the extracranial disease burden experienced by patients in their cohorts. However, we observe in our cohort that patients with extracranial metastases trended towards experiencing shorter OS but not CNS-specific PFS ( Table 2, Table 3), suggesting that at least a subset of LM patients succumb to extracranial disease rather than their LM. We do, however, observe a significant correlation between CNS-specific PFS and OS in our cohort (Supplemental Fig. S2). This suggests that CNS-specific PFS has the potential to serve as a useful surrogate endpoint for this patient population with further refinement of standardized criteria to define LM response and progression [10,21,22].
Although we were not able to obtain individual patient data from these two recent trials and were thus unable to include them in this meta-analysis, the inclusion of these data would not alter our observation that IV trastuzumab is non-inferior to IT trastuzumab. This is because the mOS of 7.9 and 10.5 months in these two studies is shorter than the mOS of 14.5 months in the group of patients who received IT trastuzumab described herein [6,7].
The mOS in our cohort of patients with LM is longer than that generally cited in the literature [1]. Moreover, in our cohort, mOS and CNS-specific mPFS are shorter for patients included in larger studies ( Table 2, Table 3). These observations can likely be attributed to publication bias, in that patients selected for publication in case reports and series experienced exceptional responses to treatment. Since this bias applies both to patients who received IT or IV treatment, it is unlikely that it would impact the key results described herein.
Beyond the lack of evidence demonstrating efficacy of IT over IV trastuzumab for HER2+ BC LM, a number of complications are associated with IT administration, such as drug-induced aseptic meningitis (DIAM) and infection of the IT reservoir through which the agents are administered. DIAM is a relatively uncommon complication of IT administration of trastuzumab, having been reported in two case reports across the literature [23,24]. However, 5%-8% of patients with an Ommaya reservoir experience Ommaya reservoir-related infections [25,26], a complication which is associated with prolonged hospital admissions and a mortality rate of approximately 10%. While the two aforementioned prospective studies demonstrate encouraging safety data for IT trastuzumab in their limited cohorts, it is clear that this approach encompasses additional risks of adverse events that are not present with IV therapy. For this reason, IT trastuzumab for HER2+ BC LM should require an additional burden of proof-of-efficacy before it is implemented outside of a clinical trial setting. While the window of opportunity for such a randomized-controlled trial has existed for several years, it is rapidly closing with novel agents such as T-DXd and tucatinib poised to be used in a growing number of HER2+ BC LM patients.
Despite the fact that we include only a small subgroup of 15 patients treated with T-DXd in this study, we were nonetheless able to observe a significant survival advantage with this agent over the rest of the population, and specifically against T-DM1 (Fig. 3). These results are in line with the recent TUXEDO-1 study demonstrating impressive efficacy of T-DXd for HER2+ BC patients with parenchymal brain metastases, and a recent publication by Alder et al. describing a case series of BC LM patients treated with T-DXd [27,28]. This sets the stage for future studies assessing the efficacy of T-DXd specifically for LM. Indeed, the ongoing DEBBRAH trial includes a cohort specifically for patients with HER2+ BC LM who will be treated with T-DXd [29]. T-DXd may have additional utility as a treatment for BC patients with HER2-low LM and HER2-mutant non-small cell lung cancer LM patients, given the positive results of the DESTINY-04 [30] and DESTINY-Lung01 trials [31].
Tucatinib is also a promising molecule under investigation for the treatment of HER2+ BC LM. We were limited in this study in that we were only able to acquire data that met inclusion criteria from 6 patients treated with tucatinib. However, preliminary data assessing the safety and efficacy of tucatinib, trastuzumab and capecitabine for HER2+ BC LM, following the positive results of this same regimen for the treatment of parenchymal brain metastases in the HER2CLIMB study, are encouraging [10,32,33].
Beyond HER2-targeted therapies, immunotherapy represents another promising treatment modality for this patient population [34,35], with IT administration of nivolumab being actively studied for LM [36]. A novel approach making use of bi-specific antibodies (HER2Bi) armed activated T-cells (HER2 BATs) was investigated in a recent phase I trial (NCT03661424). While this trial was terminated due to slow study accrual, further studies are required to assess the efficacy of HER2 BATs for the treatment of LM. Moreover, while none of the patients from our dataset received proton craniospinal irradiation, this novel RT approach has demonstrated efficacy for patients with solid tumor LM, representing another encouraging component of the treatment armamentarium for HER2+ BC LM in development [37].
There are several limitations associated with our study. Many of the patients included in this meta-analysis are derived from case reports and retrospective case series, resulting in imbalances in some of the patient characteristics between those who received IT versus non-IT therapy. While we have taken measures to control for this bias, such as performing quality assessment and performing extensive subset analyses, there is no alternative for a well-designed randomized controlled trial to directly compare HER2-targeted agents and their route of administration. In addition, we are limited by publication bias, whereby patients who experienced better than expected responses to therapy were more likely to be published in the literature. For this reason, the CNS-specific mPFS and mOS we observe herein of 6 and 14.3 months, respectively, are overestimations of the outcomes seen in real-world studies of patients with HER2+ BC LM [38]. Furthermore, data regarding CNS-specific PFS must be considered with caution as the evaluation of LM response and LM progression is highly challenging and could vary across studies [21].
Together, the results of this study demonstrate that HER2-targeted therapy is similarly active in patients with HER2+ BC LM regardless of the route of administration. T-DXd demonstrates an encouraging signal of efficacy in a small subgroup of patients. Prospective and randomized studies are warranted to define its role in the management of HER2+ BC LM.

Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgments
We dedicate this manuscript to Catherine Kargas. This study was motivated by her courage and battle against breast cancer leptomeningeal metastasis.