Identifying MAGE-A4-positive tumors for TCR T cell therapies in HLA-A∗02-eligible patients

T cell receptor (TCR) T cell therapies target tumor antigens in a human leukocyte antigen (HLA)-restricted manner. Biomarker-defined therapies require validation of assays suitable for determination of patient eligibility. For clinical trials evaluating TCR T cell therapies targeting melanoma-associated antigen A4 (MAGE-A4), screening in studies NCT02636855 and NCT04044768 assesses patient eligibility based on: (1) high-resolution HLA typing and (2) tumor MAGE-A4 testing via an immunohistochemical assay in HLA-eligible patients. The HLA/MAGE-A4 assays validation, biomarker data, and their relationship to covariates (demographics, cancer type, histopathology, tissue location) are reported here. HLA-A∗02 eligibility was 44.8% (2,959/6,606) in patients from 43 sites across North America and Europe. While HLA-A∗02:01 was the most frequent HLA-A∗02 allele, others (A∗02:02, A∗02:03, A∗02:06) considerably increased HLA eligibility in Hispanic, Black, and Asian populations. Overall, MAGE-A4 prevalence based on clinical trial enrollment was 26% (447/1,750) across 10 solid tumor types, and was highest in synovial sarcoma (70%) and lowest in gastric cancer (9%). The covariates were generally not associated with MAGE-A4 expression, except for patient age in ovarian cancer and histology in non-small cell lung cancer. This report shows the eligibility rate from biomarker screening for TCR T cell therapies and provides epidemiological data for future clinical development of MAGE-A4-targeted therapies.

Highly heterogenous tumors, such as synovial sarcoma and myxoid/round cell liposarcoma (MRCLS), require scoring in multiple high-power fields using the "field of view" method, while tumors with homogeneous MAGE-A4 expression are scored using regional method.
The percentage of MAGE-A4-positive tumor cells was determined at each intensity (0, 1+, 2+, and 3+ intensity) relative to the total number of viable tumor cells in the sample.When a MAGE-A4 signal was present in the cytoplasm and nucleus, the compartment with the highest-intensity expression was evaluated.Specific scoring rules were applied when scoring liposarcoma with myxoid regions and regions with complex arborizing vessel patterns.Arborizing vessels and the cell poor myxoid component were not taken into account when present.The high cellularity regions of the tumor were scored when scoring the myxoid liposarcoma cases.
A sample was considered MAGE-A4 positive if ≥30% tumor cells had a MAGE-A4 positivity at 2+ intensity or more (Figure S2).

Statistical analysis
A specific cutoff point (≥30% MAGE-A4-positive tumor cells stained at ≥2+ intensities) was applied on the scoring outcome to determine positivity/negativity for each sample.The positive/negative status was used to establish concordance.For precision assessment (intra-and inter-run variability), percent positive agreement and percent negative agreement of repeat staining of samples were based on the positive/negative status.The acceptance criterion for precision was set as 80% concordance at the slide level (nine replicates in three different runs for each of a minimum of four different samples per indication).Furthermore, the concordance on sample level is included for descriptive purposes.

Accuracy/specificity of the MAGE-A4 antibody
Several FFPE cell line slides and control tissue (normal testis) with known expression levels of MAGE-A4 were characterized to determine the MAGE-A4 specificity.
Cell line A375, which is known to be positive (datasheet Origene, clone OTI1F9; Sanderson et al., 2019) for MAGE-A4, demonstrated MAGE-A4-positive staining using the MAGE-A4 IHC CTA, while for cell line HCT116, known to be negative for MAGE-A4, no staining could be observed in all staining runs (Figure S3).In testis, nuclear and cytoplasmic MAGE-A4 staining was observed in the atrophic ducts and in the seminiferous tubules with strong intratubular staining, while no MAGE-A4 staining was demonstrated in the stroma (Figure S4).
No MAGE-A4 staining was observed in a normal human tissue TMA, except for human testis (Figure S5).The TMA with different (normal) human tissue types was evaluated for staining intensity and no positivity for MAGE-A4 could be observed in these normal tissues (breast, intestine, liver, lung, stomach, heart, fallopian tube) with a 10 μg/ml concentration of the primary antibody.For testis, nuclear and cytoplasmic MAGE-A4 positivity is present in the seminiferous tubules.
To further investigate the cross-reactivity of anti-MAGE-A4 antibody to MAGE-A10, three cell lines (NCI-H82, NCI-H466, and Mel526) were chosen for further characterization.By qRT-PCR, NCI-H82 was shown to have high MAGE-A4 expression and high MAGE-A10 expression (MAGE-A4 h /MAGE-A10 h ).NCI-H466 was shown to have low MAGE-A4 expression and high MAGE-A10 expression (MAGE-A4 l /MAGE-A10 h ).Mel526 was shown to have negligible MAGE-A4 expression and high MAGE-A10 expression (MAGE-A4 n /MAGE-A10 h ) (Figure S6C).Anti-MAGE-A4 antibody showed strongly positive, weakly positive and negative staining in NCI-H82, NCI-H466, and Mel526, respectively, supporting the specificity of anti-MAGE-A4 antibody for MAGE-A4 without cross-reactivity to MAGE-A10 (Figure S6D).Some low-intensity staining could be observed with the anti-MAGE-A4 antibody in transduced NALM6 cells expressing extremely high levels of MAGE-A10 (Figure S6B), indicating a possible low-affinity crossreactivity that can only be detected in this artificial system, which may not be physiologically or pathologically relevant.
Further evidence of the specificity of the MAGE-A4 CTA and estimation of the potential impact of the cross-reactivity with MAGE-A10 on its diagnostic value was provided by the analysis of data from 352 tumor samples from 10 different indications (melanoma, bladder cancer, NSCLC, head and neck cancer, esophageal cancer, esophagogastric junction cancer, ovarian cancer, gastric cancer, MRCLS, synovial sarcoma) stained for expression of both MAGE-A4 and MAGE-A10.These samples were screened under a screening protocol (ADP-0000-001, NCT02636855) used to determine eligibility for enrollment into one of two clinical trials using T cells directed against MAGE-A10 (ADP-0022-003, NCT02592577 and ADP-0022-004, NCT02989064) as well as a clinical trial using T cells directed against MAGE-A4 (ADP-0044-001, NCT03132922).For detection of MAGE-A10, a CTA based on a goat polyclonal antibody (Santa Cruz, Cat # sc-324906) was developed, validated, and used under a CLIA-certified laboratory to stain by IHC sections from FFPE tumor samples.Serial sections from the same samples were stained with the MAGE-A10 CTA.Similar to the MAGE-A4 CTA, scoring for the MAGE-A10 CTA was based on the percentage of live tumor cells stained at intensities of 0, 1+, 2+, or 3+.Most tested samples showed no expression of either target proteins.Figure S6E shows the P score (percentage of tumor cells stained at ≥2+) for both MAGE-A4 and MAGE-A10 in a selection of positively stained samples.Among the positively stained samples, the majority had expression of both MAGE-A4 and MAGE-A10 with similar levels.Four samples showed MAGE-A4 staining with a P score of 100 (with 3+ staining intensity in 70%-100% of tumor cells) but MAGE-A10 P score of 0, demonstrating the specificity of the anti-MAGE-A10 antibody, without cross-reactivity to MAGE-A4 (all the four data points overlapped and are shown as one in Figure S6E).Conversely, eight samples (Figure S6E, circled) showed P scores for MAGE-A10 between 20 and 100 (with 3+ staining intensity in 10%-70% of tumor cells) and no or very low staining for MAGE-A4 (P scores between 0 and 10).The low staining for MAGE-A4 observed in these samples could be due to actual low expression of MAGE-A4, but, even assuming that the MAGE-A4 signal is entirely due to crossreactivity of the anti-MAGE-A4 antibody with MAGE-A10, this low cross-reactivity would not change the MAGE-A4 diagnosis status of these samples (positivity cutoff, ≥30% ≥2+), thus negating the risk of false positivity and confirming the diagnostic validity of the MAGE-A4 CTA.
In addition, anti-MAGE-A4 antibody staining by IHC showed no staining in Mel624 cell line, which is MAGE-A11 positive and MAGE-A12 positive by mRNA profile (Figure S6D).This further indicates the specificity of anti-MAGE-A4 antibody without cross-reactivity to MAGE-A11 and MAGE-A12.S1).For the determination of the prevalence of MAGE-A4 in different tumor indications, a MAGE-A4 cutoff of ≥30% at a ≥2+ intensity was applied.Feasibility of the MAGE-A4 assay was assessed in 316 tissue samples distributed over nine tumor indications covering the complete MAGE-A4 dynamic range (Figure S2, Table S1).Of the 316 tissue samples tested, 80 samples were positive for MAGE-A4.As demonstrated in Figure S2, the prevalence of MAGE-A4 ranged from 6% in EGJ cancer up to 67% in synovial sarcoma.Lower prevalence is observed in gastric cancer, EGJ cancer, MRCLS, ovarian cancer, and endometrium carcinoma, while in synovial sarcoma, HNSCC, and esophageal squamous cell carcinoma, and adenocarcinoma the prevalence is higher.

Robustness of the assay
To evaluate the robustness of the MAGE-A4 IHC CTA, precision testing (intra-run and inter-run), inter-lot (lot 1 vs. lot 2) and inter-lab (CellCarta BE vs. CellCarta US) comparison was evaluated to confirm the MAGE-A4 IHC CTA robustness regardless of the antibody lot used or the lab performing the MAGE-A4 assay.Each sample tested was stained for MAGE-A4 and their corresponding isotype control.
The MAGE-A4 IHC CTA robustness was evaluated both qualitatively and semi-quantitatively as scored by a pathologist.For each slide, the MAGE-A4 status (positivity cutoff: ≥30% tumor cells stained by MAGE-A4 at a ≥2+ intensity) was determined.The evaluation was based on the case status of each sample.All serial sections of each sample should be positive or negative in all the reads and an 80% overall concordance (overall percent agreement [OPA]) must be reached on slide level to consider the robustness as valid.The robustness results are summarized below.

Repeatability and reproducibility: precision
To evaluate the robustness (intra-run and inter-run) of the MAGE-A4 IHC assay, precision testing was performed in three independent staining runs on non-consecutive days on at least two Dako Link autostainer platforms by at least two different operators (Figure S7) to evaluate inter-run, intra-run, interoperator, and inter-instrument variability.In each run, four serial sections were stained (three slides with the positive protocol and one with the negative protocol) from each block to evaluate the repeatability (intra-run) and reproducibility (inter-run) of the assay.Over three runs, 12 slides (nine with positive protocol and three with negative protocol) were stained per block.
Based on these results from the qualitative (Figures S8 and S9) and semi-quantitative (Table S2) evaluation, the precision of the MAGE-A4 assay was confirmed and each indication tested showed a concordance of >80%.Therefore, the MAGE-A4 assay was considered robust.Since different operators and instruments were used, the MAGE A4 assay is robust regardless of the operator or Dako Link autostainer instrument used for staining.MRCLS showed the lowest robustness (OPA 89%) since five slides from two samples deviated resulting in a different category.In the first sample, one slide was scored negative (22% at ≥2+ intensity), whereas the average score for this samples was 37% at ≥2+ intensity.In the second sample, four slides were scored negative (21%, 27%, 29%, and 29% at ≥2+ intensity), whereas the average score for this sample was 33% at ≥2+ intensity.Since the cutoff for MAGE-A4 is 30%, both samples were borderline cases, and all slides were scored around the cutoff.The variation on slide level was minimal (15% CV and 20% CV, respectively) but since categorization was used, this resulted in a different category.

Inter-lab variability
The MAGE-A4 assay was initially validated at CellCarta Antwerp (CC BE).After validation at CC BE, the MAGE-A4 assay was transferred to CellCarta Naperville (CC US).All of CellCarta's laboratories are CAP/CLIA certified.All staining platforms at all sites are cross validated twice per year.
For inter-lab variability between CellCarta Antwerp (CC BE) and CellCarta Naperville (CC US), two serial slides of 12 tissue samples (six sarcoma and six carcinoma samples) (Table S3) were stained at both labs and evaluated for concordance using the ≥30% positivity at a ≥2+ intensity cutoff.
As demonstrated in Table S3, one sample, a MRCLS, had a higher variability (CV 47% [2.53/5.36])compared to the other samples (CV <10%) although the MAGE-A4 status remained unchanged.Biological variation between the two stained slides could lead to the observed variability of staining.
Based on the results, it has been concluded that the MAGE-A4 assay performs similarly regardless of the staining lab, CC BE or CC US.The robustness of the MAGE-A4 assay is therefore confirmed.Representative images of the inter-lab comparison are included in Figures S10-S13.
The results of inter-lot variability are presented in Figures S14-S16 and Table S4.In general, F001 showed a slightly weaker staining compared to A001 and F002.However, the scoring was not affected, and the lot-to-lot variability was considered valid.
Based on the results, it has been concluded that the MAGE-A4 assay performs similarly regardless of the MAGE-A4 antibody lot used for testing.No immunoreactivity is detected in the negative IgG controls.IgG, immunoglobulin; MAGE-A4, melanoma-associated antigen A4.

Figure S8 .
Figure S8.Representative detail images of the precision assessment of MAGE-A4 on a negative ovarian cancer tissue with 20% MAGE-A4 positivity at ≥2+ intensity during sensitivity screening.Three serial sections of each sample were stained in each run for MAGE-A4.In each run, one slide was stained for the isotype IgG control.Repeatability was assessed in run 1, reproducibility was evaluated between the three different runs.10x magnification.IgG, immunoglobulin G; MAGE-A4, melanomaassociated antigen A4.

Figure S9 .
Figure S9.Representative detail images of the precision assessment of MAGE-A4 on a positive endometrium cancer tissue with a 94% MAGE-A4 positivity at ≥2+ intensity during sensitivity screening.Three serial sections of each sample were stained in each run for MAGE-A4.In each run, one slide was stained for the isotype IgG control.Repeatability was assessed in run 1, reproducibility was evaluated between the three different runs.20x magnification.IgG, immunoglobulin G; MAGE-A4, melanomaassociated antigen A4.

Table S5 . Prevalence of HLA-A*02 alleles in different races and ethnicities in patients screened by the NMDP. 1,2
API, Asian or Pacific Islander; NMDP, National Marrow Donnor Program.Numbers represent the percentage of individuals expressing each allele or group of alleles in the population of interest using the formula Pi = 2 x F -(F 2 ), where Pi is the percentage of individuals expressing the allele and F is the allele frequency.