Quantitative Measurement of HER2 Expression in Non-Small Cell Lung Cancer with a High Sensitivity Assay

Abstract

Recently, low HER2 protein expression has been proposed as a predictive biomarker for response to antibody-drug conjugate trastuzumab deruxtecan (T-DXd) in metastatic breast cancer. HER2 expression in non-small cell lung cancer (NSCLC) patients has never been carefully measured, and little is known about the frequency of cases with unamplified but detectable levels of the protein. Although some HER2-targeted therapies have been studied in NSCLC patients, they have been restricted to those with genomic ERBB2 gene alterations, which only represent relatively rare cases of NSCLC. Still, emerging investigations of T-DXd in NSCLC have shown promise in patients with unamplified HER2. Taken together, we hypothesize that there may be many cases of NSCLC with levels of HER2 protein expression comparable to levels seen in breast cancer who benefit from T-DXd.

Here, we used a previously validated, analytic, quantitative immunofluorescence (QIF) assay that is more sensitive than legacy clinical HER2 immunohistochemistry assays. We measured HER2 protein levels in NSCLC cases to determine the proportion of cases with detectable HER2 expression. Using cell line calibration microarrays alongside our QIF method enabled us to convert HER2 signal into units of attomoles per mm². We found that over 63% of the 741 analyzed NSCLC cases exhibited HER2 expression above the limit of detection, with more than 17% of them exceeding the lower limit of quantification. While the threshold for response to T-DXd in breast cancer is still unknown, many cases of NSCLC have expression in a range comparable to breast cancer cases with immunohistochemistry scores of 1+ or 2+. Our assay could potentially select NSCLC cases with detectable target (i.e., HER2) that might benefit from HER2 antibody-drug conjugates, irrespective of ERBB2 genomic alterations.

INTRODUCTION

Human epidermal growth factor receptor 2 (HER2), the common name for the protein coded by the ERBB2 gene, is among the most impactful therapeutic targets. Until recently, HER2-targeted therapies have been successful only in cases with protein overexpression and/or gene amplification (predominantly breast and gastric cancer) ^1,2. However, the development of antibody-drug conjugates using HER2 as a protein tag on cancer cells has changed the landscape. Trastuzumab deruxtecan (T-DXd), the trastuzumab antibody conjugated with a topoisomerase inhibitor ³, has shown great promise in breast cancer, first in the amplified advanced stage cancers ⁴, but then more recently in advanced cancer with expression of HER2 in the low range in the absence of ERBB2 gene amplification ⁵. More recent work has led to the FDA approval of T-DXd in advanced breast cancer in patients with HER2 immunohistochemistry (IHC) scores of 1+ and 2+ in the absence of gene amplification ⁶.

ERBB2 alterations, which include ERBB2 gene mutation, ERBB2 gene amplification, and HER2 protein overexpression ^7,8, are relatively rare in lung cancer. These alterations have spurred clinical trials of HER2-targeted therapies in NSCLC, albeit not to the extent seen in other tumor types. ERBB2 mutation and ERBB2 gene amplification in NSCLC make up only 2–5% of all cases ^9–13. But ERBB2-mutated NSCLC has predominantly been targeted in T-DXd clinical trials ¹⁴. A very promising clinical response (55% objective response rate) was seen in patients with ERBB2 mutation independent of IHC-assessed protein expression ¹⁵. HER2 protein has been assessed in NSCLC, and overexpression is generally low, although rare studies have reported frequencies above 15% ^16–18. However, the IHC assays for these studies are the same as those used in breast cancer, where the IHC test is calibrated to distinguish very high levels of protein expression seen in gene-amplified cases from cases without gene amplification. As such, the assay can distinguish millions of copies of the protein per cell from hundreds of thousands but is insensitive in lower ranges of protein expression. A number of studies have suggested that HER2 has a broad dynamic range of expression even in unamplified breast cancer ^19,20, raising the question of the dynamic range of expression in NSCLC. Given the recent success of T-DXd in low HER2-expressing breast cancer, expression of HER2 in the “low-HER2” range in NSCLC is now a provocative open question.

Since the current most commonly used IHC assay for HER2 in breast cancer has been described as weighing mice on a scale designed for elephants ²¹, we developed an assay centered in the proper dynamic range for HER2 expression in gene-unamplified breast cancer ²². By using cell lines with known levels of expression of HER2 as a standard curve, we are able to assess low levels of HER2 in units of attomoles (amol)/mm². This assay, called High Sensitivity HER2 (HS-HER2), is now offered in our CLIA lab. We have now begun efforts to determine the levels of HER2 that are associated with response to T-DXd in amol/mm².

In this study, we performed quantitative immunofluorescence (QIF) to measure HER2 expression on retrospective collections of NSCLC cases using the HS-HER2 assay. By using calibration arrays as part of the assay to determine HER2 protein amol/mm², here we define the expression of HER2 in NSCLC cases using analytic parameters, including the limit of detection (LOD), the limit of quantification (LOQ), and the limit of linearity (LOL).

MATERIALS AND METHODS

HER2 calibration CMA

As previously described ²², the HER2 calibration cell line microarray (CMA) was built by Array Science LLC (Sausalito, CA) containing a two-fold redundancy of the cell lines JURKAT, BT-20, T-47D, ZR-75-1, BT-483, AU565, and BT-474.

NSCLC cohorts (YTMA399, YTMA423, YTMA553)

All tissue specimens were prepared as tissue microarrays (TMAs) from 0.6 mm cores of formalin-fixed, paraffin-embedded tissue. We used three retrospective collections of lung carcinoma (YTMA 399, YTMA 423, and YTMA 533), containing 172, 287, and 296 NSCLC cases collected from 2011–2017, 2011–2016, and 2017–2020, respectively. Collections from YTMA 399 were from the National and Kapodistrian University of Athens, while those from YTMA 423 and YTMA 553 were from Yale University. Basic clinicopathologic characteristics of the individual cohorts are outlined in Table 1. For QIF on the cohorts, we used replicate slides (n = 6 for YTMA 399, n = 6 for YTMA 423, and n = 5 for YTMA 553) across three blocks cored from different regions of tumors. Scores obtained from replicate spots and cases were averaged for analyses.

Table 1:

Patient characteristics across three YTMA cohorts

Characteristic	YTMA 399 (N = 172)	YTMA 423 (N = 287)	YTMA 553 (N = 296)
Sex
Female	34 (19.8%)	175 (61.0%)	167 (56.4%)
Male	116 (67.4%)	112 (39.0%)	128 (43.2 %)
NA	22 (12.8%)	0 (0%)	1 (0.3%)
Median age, years (range)	65 (37–82)	68 (38–89)	68 (32–89)
Race
White	0 (0%)	267 (93.0%)	0 (0%)
Black	0 (0%)	15 (5.2%)	0 (0%)
Asian	0 (0%)	4 (1.4%)	0 (0%)
Other	0 (0%)	1 (0.3%)	0 (0%)
NA	172 (100%)	0 (0%)	296 (100%)
Smoke status
Never	10 (5.8%)	40 (13.9%)	1 (0.3%)
Former	81 (47.1%)	180 (62.7%)	248 (83.8%)
Current	57 (33.1%)	67 (23.3%)	45 (15.2%)
NA	24 (14.0%)	0 (0%)	2 (0.7%)
Performance status
0	73 (42.4%)	238 (82.9%)	0 (0%)
1–2	69 (40.1%)	46 (16.0%)	0 (0%)
3–4	6 (3.5%)	3 (1.0%)	0 (0%)
NA	24 (14.0%)	0 (0%)	296 (100%)
PD-L1 tumor expression
Negative (<1%)	5 (2.9%)	143 (48.8%)	98 (33.1%)
1–49%	2 (1.2%)	52 (18.1%)	57 (19.3%)
≥ 50%	2 (1.2%)	37 (12.9%)	34 (11.5%)
NA	163 (94.8%)	55 (19.2%)	107 (36.1%)
Stage at diagnosis
I	38 (22.1%)	200 (69.7%)	214 (72.3%)
II	43 (25.0%)	66 (23.0%)	47 (15.9%)
III	48 (27.9%)	17 (5.9%)	23 (7.8%)
IV	21 (12.2%)	1 (0.3%)	1 (0.3%)
NA	22 (12.8%)	3 (1.0%)	11 (3.7%)
Histology
Adenocarcinoma	94 (54.7%)	206 (71.8%)	221 (74.7%)
Squamous cell	49 (28.5%)	67 (23.3%)	53 (17.9%)
Adenosquamous	4 (2.3%)	1 (0.3%)	0 (0%)
Carcinoid	0 (0%)	0 (0%)	12 (4.1%)
Large cell	0 (0%)	7 (2.4%)	1 (0.3%)
Other/NOS	25 (14.5%)	6 (2.1%)	9 (3.0%)
Outcome from last follow-up
No progression/recurrence	72 (41.9%)	218 (76.0%)	252 (85.1%)
Progression/recurrence	78 (45.3%)	46 (16.0%)	39 (13.2%)
NA	22 (12.8%)	23 (8.0%)	5 (1.7%)
Vital status from last follow-up
Alive	35 (20.3%)	212 (73.9%)	252 (85.1%)
Dead	115 (66.9%)	70 (24.4%)	40 (13.5%)
NA	22 (12.8%)	5 (1.7%)	4 (1.4%)
Survival, months (95% CI)
Median progression/recurrence free survival	15.1 (10.2–19.1)	48.0 (41.0–50.0)	31.1 (28.2–36.0)
Median overall survival	28.5 (16.5–38.8)	49.5 (46.0–53.0)	33.0 (29.0–36.3)

Note: Progression pertains to YTMA 399 and YTMA 423 while recurrence pertains to YTMA 553.

Abbreviations: CI, confidence interval; NA, not available; NOS, not otherwise specified; PD-L1, programmed death-ligand 1; YTMA, Yale tissue microarray.

Antibodies and quantitative immunofluorescence

Slides were deparaffinized and subjected to antigen retrieval with EDTA pH8 buffer at 97 °C for 20 minutes in a pressurized heating device (PT module, Lab Vision). The slides were then incubated for 30 minutes in 0.3% hydrogen peroxide in methanol to inactivate endogenous peroxidases prior to overnight incubation in a blocking solution of 0.3% bovine serum albumin and 0.05% tween-20. Primary antibodies of anti-HER2 (clone 29D8, Cell Signaling) at a concentration of 1 μg/ml (1:256 dilution) and anti-cytokeratin (CK) (clone AE1/AE3, Dako) at a concentration of 1.675 μg/ml (1:100 dilution) were applied on the tissue for 1 hour. The two antibodies were validated in our lab and previously published ^22,23. A subsequent 1-hour incubation of anti-rabbit Envision (Dako) and anti-mouse Alexa Fluor 546 (Invitrogen, 1:100) secondary antibodies was performed, followed by the addition of Cyanine 5 tyramide (Akoya Biosciences, 1:50) to amplify HER2 signal. 4’,6-diamidino-2-phenylindole (DAPI) was used to stain nuclei, and the slides were mounted with ProLong Gold Antifade reagent (Invitrogen). Images were acquired on the PM-2000 system (Navigate Biopharm, Carlsbad, CA) and analyzed under the AQUA^™ method of QIF, as previously described (Supplementary Figure 1) ²⁴. HER2 target was measured within a tumor mask created by binarizing and dilating the CK signal. All TMA spots were visually assessed, and those with staining artifacts or less than 2% of compartment area were excluded from analyses.

Staining of the NSCLC TMAs was performed in batches with at least one CMA per batch. HER2 QIF scores from the NSCLC cases were converted to units of amol/mm² using calibration curves from CMAs associated with the same staining batch.

Chromogenic staining

Chromogenic staining served as a secondary, qualitative approach to HER2 visualization. Slides underwent the same conditions as the QIF staining protocol up to the overnight blocking step, after which they were subjected to a 1-hour incubation of the anti-HER2 antibody at 1 μg/mL (1:256 dilution) and a 1-hour incubation of anti-rabbit Envision (Dako). 3,3’-diaminobenzidine (DAB) in the form of Liquid DAB+ Substrate Chromogen (Dako) was then used, followed by counterstaining with Dako hematoxylin. The slides were finally dehydrated in ethanol and xylene, mounted with Cytoseal 60 (Thermo Scientific), and scanned using the Aperio AT2 (Leica Biosystems). To chromogenically compare the standard assay to the high sensitivity assay, two CMAs and two 553 TMAs were processed as follows: One CMA-TMA pair underwent the same protocol but with an anti-HER2 antibody concentration of 3 μg/mL (1:85.3 dilution). The other pair used a concentration of 0.03 μg/mL (1:8533 dilution), emulating the staining seen in standard HER2 IHC assays

Calculation of limits of detection, quantification, and linearity

To determine the limit of detection (LOD) and limit of quantification (LOQ) for the QIF HS-HER2 assay, we used a signal-to-noise approach in accordance with recommendations from the FDA Guidance for Industry Q2 (R1) ²⁵. Briefly, we plotted the signal response of the calibration CMA spots against the analyte concentration (amol/mm²), which were obtained previously ²² for JURKAT (0 amol/mm²), BT-20 (6.6 amol/mm²), T-47D (12.9 amol/mm²), ZR-75-1 (18.1 amol/mm²), and BT-483 (21.8 amol/mm²) cell lines (Supplementary Figure 2A). We had a total of 5 CMA slides derived from the QIF experiments we performed, amounting to a 10-fold redundancy of cell lines. A regression analysis was performed on the pooled data, whereby the average signal response of JURKAT (defined as HER2-negative) was multiplied by three or ten depending on whether the LOD or LOQ was being determined, per FDA guidance. The multiplied values were then incorporated into the regression equation to obtain the respective analyte concentrations corresponding to the LOD and LOQ.

To determine the limit of linearity (LOL), as described by Huber ²⁶, we looked at the response factors for each method of HER2 quantification (Supplementary Figure 2B). The greater analyte concentration where the regression line intersects with lines y = 1.05α or y = 0.95α, where α is the average of the response factors, would be defined as the upper LOL (Supplementary Figure 2B). Mentions of LOL in this paper will specifically refer to the upper LOL.

Statistical analysis

Statistical analysis was carried out using GraphPad Prism 9. Linear regression analyses were performed using Pearson’s correlation coefficient. Among those analyses, QIF scores from serial sections of separate experimental runs were compared, yielding R² values above 0.8, which confirmed the overall reproducibility of the HS-HER2 assay (Supplementary Figures 3A–C). Comparisons of distributions stratified by patient characteristics were performed using the Mann-Whitney test or the Kruskal-Wallis one-way ANOVA test followed by Dunn’s test. Survival analysis was visualized using Kaplan-Meier curves, with significance determined by the log-rank test. All reported p-values were not adjusted for multiple comparisons beyond any corrections resulting from Dunn’s test as a follow-up to ANOVA.

RESULTS

Determining the LOD, LOQ, and LOL from 5 CMAs

For an analytic assay, the LOD is the lowest detectable signal that can be sufficiently distinguished from noise, while the LOQ is the lowest signal that can be quantified with a reliable degree of precision ²⁷. The LOL is the signal level beyond which the calibration curve is no longer linear ²⁶. Using pooled QIF data from the 5 stained CMAs, as explained in the methods, resulted in calculations of 1.5 amol/mm² for the LOD, 4.4 amol/mm² for the LOQ, and 22.3 amol/mm² for the LOL. To test the soundness of our LOD and LOQ values, we compared pairs of tissue serial section sets from each of the three cohorts. We observed noticeably improved regression between cases in the LOQ to LOL range (R² approx. 0.8) compared to cases in the LOD to LOQ range (R² approx. 0.6) (Supplementary Figures 4A–D). It is notable that the reproducibility is still quite high even in the range between the LOD and LOQ.

Frequency of HER2 expression in 741 NSCLC cases

Using the LOD, LOQ, and LOL values we had derived, we assessed expression of HER2 in the cytokeratin compartment of all NSCLC cases averaged across all replicates. Of the 741 evaluable cases representing the combination of 3 serially collected cohorts from two separate institutions, 342 (46.2%) were between the LOD and LOQ, 126 (17.0%) were between the LOQ and LOL, and 4 (0.54%) were above the LOL (Figure 1). Representative images of HER2 QIF staining in the NSCLC cases are shown in Figures 2A–D. HER2 DAB staining of the same cases from serial cuts is shown in Figures 2E–H.

Figure 1.

Ordered bar graph of 741 NSCLC cases from YTMA 399 (6-fold redundancy), YTMA 423 (6-fold redundancy), and YTMA 553 (5-fold redundancy), measured in averaged amol/mm². LOD, LOQ, and LOL cutoffs are indicated. The graph’s maximum height is set at 30 amol/mm² to emphasize the majority of the distribution; three cases surpassing 30 amol/mm², with values of 42.8 amol/mm², 46.1 amol/mm², and 53.5 amol/mm², are truncated.

Figure 2.

Representative images of differential HER2 expression in NSCLC cases. Representative IF staining of HER2 in NSCLC cases is observed (A) below the 1.5 amol/mm² LOD (0.84 amol/mm² for this case), (B) above the 1.5 amol/mm² LOD but below the 4.4 amol/mm² LOQ (2.9 amol/mm² for this case), (C) above the 4.4 amol/mm² LOQ but below the 22.3 amol/mm² LOL (16.2 amol/mm² for this case), and (D) above the 22.3 amol/mm² LOL (40.5 amol/mm² for this case). Pseudocolor shows tumor (CK) in green, HER2 in red, and nuclei (DAPI) in blue. Select regions are shown at higher magnification. High-power images of HER2 DAB staining in (E), (F), (G), and (H) correspond to serial sections of cases seen in (A), (B), (C), and (D), respectively. Scale bars: 50 μm.

Frequency of HER2 expression by characteristic

We then examined potential differences in mean HER2 levels by specific patient characteristics (Figures 3A–F). When stratifying by histology, we found that adenocarcinoma cases had significantly higher HER2 levels than squamous cell carcinoma cases (p < 0.0001) (Figure 3A). When stratifying by sex, we found that female patients had significantly higher HER2 levels than male patients (p = 0.0307) (Figure 3B). When stratifying by tumor PD-L1 status, we found that cases with low (0%–49%) levels of PD-L1 had significantly higher HER2 levels than cases with high (≥50%) levels of PD-L1 (p = 0.0042) (Figure 3C). We additionally stratified cases by stage (I–II, III, and IV), race (white and black), and smoking status (current, former, and never) but observed no significant differences in HER2 levels except when we compared current smokers to former smokers (p = 0.0456) (Figures 3D–F). Table 2 summarizes all the stratified cases by LOD, LOQ, and LOL cutoffs.

Figure 3.

HER2 levels in NSCLC cases stratified by (A) histology (adenocarcinoma, n = 514; and squamous cell carcinoma, n = 168), (B) sex (male, n = 346; and female, n = 373), (C) stage at diagnosis (I–II, n = 600; III, n = 87; and IV, n = 19), (D) tumor PD-L1 status (0%–49%, n = 353; and ≥50%, n = 73), (E) race (white, n = 263; and black, n = 15), and (G) smoking status (current smoker, n = 149; former smoker, n = 472; and never-smoker, n = 96). Mean with 95% confidence intervals are indicated for each group. Y-axes are broken where appropriate to emphasize the majority of the distribution. Adeno, adenocarcinoma.

Table 2:

Proportion of stratified NSCLC cases with HER2 levels below the LOD, between the LOD and LOQ, between the LOQ and LOL, and above the LOL

Characteristic	Below LOD	LOD to LOQ	LOQ to LOL	Above LOL
Histology
Adenocarcinoma (n = 514)	129 (25.1%)	267 (51.9%)	115 (22.4%)	3 (0.6%)
Squamous (n = 167)	106 (63.5%)	56 (33.5%)	4 (2.4%)	1 (0.6%)
Stage
I–II (n = 600)	215 (35.8%)	272 (45.3%)	110 (18.3%)	3 (0.5%)
III (n = 87)	35 (40.2%)	44 (50.6%)	8 (9.2%)	0 (0%)
IV (n = 19)	4 (21.1%)	11 (57.9%)	3 (15.8%)	1 (5.3%)
Sex
Male (n = 346)	134 (38.7%)	154 (44.5%)	57 (16.5%)	1 (0.3%)
Female (n = 373)	124 (33.2%)	180 (48.3%)	66 (17.7%)	3 (0.8%)
Tumor PD-L1
0%–49% (n = 353)	110 (31.2%)	171 (48.4%)	69 (19.5%)	3 (0.8%)
≥50% (n = 73)	33 (45.2%)	33 (45.2%)	7 (9.6%)	0 (0%)
Race
White (n = 263)	101 (38.4%)	125 (47.5%)	35 (13.3%)	2 (0.8%)
Black (n = 15)	7 (46.7%)	7 (46.7%)	1 (6.7%)	0 (0%)
Smoking Status
Current (n = 149)	64 (43.0%)	70 (47.0%)	15 (10.1%)	0 (0%)
Former (n = 472)	163 (34.5%)	219 (46.4%)	87 (18.4%)	3 (0.6%)
Never (n = 96)	31 (32.3%)	43 (44.8%)	21 (21.9%)	1 (1.0%)

Note: Total subgroups do not add to all 741 evaluable cases due to incomplete clinical information and non-exhaustive inclusion of subgroups for certain stratifications.

Abbreviations: PD-L1, programmed death-ligand 1; WT, wild-type.

Association between HER2 expression and outcome in NSCLC

Although some studies suggest a slightly worse outcome for NSCLC patients carrying HER2 alterations than those without ^16,28, no clear association has been described ²⁹. We thus explored the prognostic value of HER2 expression in our 741 cases, stratifying them by cohort since outcome information was not identical in all cohorts. Setting the cut-point at the LOQ, we found no significant association between HER2 levels and progression-free survival in YTMA 399 and 423 (Figures 4A and B), recurrence-free survival in YTMA 553 (Figure 4C), or overall survival in any of the cohorts (Figures 4D–F).

Figure 4.

Association of HER2 expression in NSCLC with survival. Shown are PFS in (A) YTMA 399 and (B) YTMA 423; RFS in (C) YTMA 553; and OS in (D) YTMA 399, (E) YTMA 423, and (F) YTMA 553. Dotted lines represent 95% confidence intervals for the survival curves. The cut-point is defined by the LOQ. OS, overall survival; PFS, progression-free survival; RFS, recurrence-free survival.

IHC comparison of the HS- HER2 assay to standard chromogenic HER2 assays.

The legacy and currently FDA approved HER2 assay (Ventana 4B5) was designed to separate ERBB2 amplified tumors from non-amplified tumors in an easy to interpret binary manner. As such it is not optimally designed to assess HER2 protein expression at low levels as seen in unamplified tumors. The new HS-HER2 assay is centered on a lower dynamic range that stratifies unamplified tumors and allows the amplified tumors to saturate the assay. This difference in assay sensitivity can alter the reading of tumors with low levels of HER2 proteins as illustrated in supplementary figure 5.

DISCUSSION

Prior assessments of HER2 in NSCLC have focused mainly on alterations including gene mutation, gene amplification, and protein overexpression. Attempts to investigate overall HER2 expression in NSCLC have been limited to low-sensitivity assays designed for breast cancer ^16,30, compounded by the lack of a standardized assay with the appropriate dynamic range that would enable a reliable level of quantification. Our effort is the first to examine HER2 expression in NSCLC with a sensitivity and resolution higher than the standard HER2 IHC assays used in the clinic and approved for cancers such as breast cancer ^22,31. In our study, we demonstrated the applicability of our previously developed HS-HER2 assay in NSCLC. We then showed in our combined cohort set that over 63% of NSCLC cases have HER2 levels above the LOD, with over 17% of them above the LOQ (Figure 1). This finding is quite high compared to the relatively low rate of HER2 genomic alterations in NSCLC ³², demonstrating that many cases may express the protein without gene mutation or amplification.

Stratification of HER2 expression by various clinicopathologic factors shows some provocative correlations. Notably, adenocarcinoma cases had significantly higher levels of HER2 than squamous cell carcinoma cases (Figure 3A), the former with a cumulative 74.9% above the LOD and 23.0% above the LOQ (Table 2). This trend is consistent with previous studies noting a higher incidence of HER2 alterations in NSCLC adenocarcinomas compared to squamous cell carcinomas ^33,34. We also found that female patients were more likely to have higher levels of HER2 than male patients (Figure 3B), which could be related to findings showing that female patients have higher rates of ERBB2 alterations ^35,36. Similarly, our observation of PD-L1-low cases having significantly higher levels of HER2 than PD-L1-high cases (Figure 3C) is in line with studies that have reported a negative correlation between PD-L1 status and ERBB2 mutation ^37,38.

The relatively high frequency of HER2 expression in NSCLC makes it a provocative biomarker for further study. Although we did not find a prognostic association between HER2 and outcome in any of the three cohorts (Figures 4A–F), recent studies have highlighted a predictive role HER2 may have in NSCLC treated with targeted therapies ^15,39. Since HER2 antibody-drug conjugates illustrate the value of low-expressing HER2 as a predictive biomarker, it will be interesting to see if the finding of the predictive value of low HER2 in breast cancer can be extended to NSCLC cases.

Our study presents several limitations. The most significant limitation is our inability to assess genomic alteration status as a function of HER2 protein expression. This is an important limitation, but not addressable on this study based on archival tissues tested in TMAs. Another related significant weakness is our reliance on retrospective tissue collected from 2 institutions. While our sample size is not small, we are underpowered to assess certain variables. Interpretation of Figure 3, which showcases preliminary stratification of some variables, warrants caution due to its early-stage analysis and lack of adjustment for multiple testing. The borderline association seen in Figures 3F would most likely lose significance after p-value correction. Also, we measured tissue on TMAs, and although we always average multiple cores of the block, we may not have adequately addressed the heterogeneity of HER2 expression, a phenomenon not only well-known in breast cancer but also implicated in NSCLC ^40,41. Indeed, we noted that HER2 levels regressed more poorly across blocks than across serial sections for our replicate slides (Supplementary Figures 3A–C). Future studies are needed using whole tissue specimens to determine the accuracy of averaged values and the importance of understanding heterogeneity, especially in the context of a predictive marker. However, these studies are best done in the context of a prospective clinical trial. Other limitations include the lack of control of preanalytical variables associated with each TMA block and tissue age of the tissue in the cohorts. Finally, we do not know the level of HER2 expression (in molecules per cell or attomole/mm²) required for response to T-DXd or similar drugs in either breast cancer or lung cancer. Follow-up studies are pivotal to determining the HER2 threshold for drug responses. In the present study, we show both the LOD and the LOQ to give a broad view of the expression levels. It is possible that even more sensitive assays will be needed once the threshold for targeted therapy benefit is established.

In conclusion, we found a substantial proportion of NSCLC cases with detectable and quantifiable levels of HER2 expression. We look forward to assessing the HS-HER2 assay for clinical utility in the prediction of response to antibody-drug conjugates for NSCLC patients.

DATA AVAILABILITY

The data generated in this study are available upon request from the corresponding author.