The use of different expensive autostainers and various assays is neither economical nor reasonable for pathology laboratories. Interchangeability of different assays may enable the usage of only one standardized PD-L1 assay in laboratories. In NSCLC, Adam et al. [16] showed that 28-8, 22C3, and SP263 assays demonstrated close analytical performance for TC staining across seven centers. However, a significant discrepancy was observed between SP142 and the other three assays for TC staining, whereas IC staining results were similar [17–19]. Moreover, the SP142 assay was an outlier that detected markedly less PD-L1 expression in both TCs and ICs [5].

Rijnders et al. [3] have suggested that agreement in the PD-L1 status in UC between 22C3, 28-8, SP142, and SP263 is substantial (80%–90%), implying that these assays may be interchangeable in clinical practice. Moreover, a collaborative study conducted by the Russian Society of Clinical Oncology and the Russian Society of Pathology found that a patient with UC classified as negative by one of the three tests (22C3, SP142, and SP263) using the corresponding cutoff rule was highly likely (91%–100%) to be classified as negative based on the results of any other test performed, therefore avoiding the need for repeated testing [8]. Furthermore, Zajac et al. [9] reported a high level of analytical concordance among the SP263, 22C3, and 28-8 assays for TC and IC staining; however, such a level of concordance was not observed for SP142. Additionally, Hodgson et al. [7] demonstrated that SP142 TC staining intensity was lower in UC and hypopharyngeal squamous cell carcinoma samples, although there existed adequate analytic comparability between 22C3 and SP263. Another recent study has highlighted greater differences in the assays used for the analysis of PD-L1–stained TCs, particularly between SP142 and other assays [20]. These analytical findings were consistent with other studies conducted using NSCLC samples, which suggested that SP142 did not exhibit sufficient concordance with the other three assays. While 22C3 and SP263 have a high concordance rate and can be used interchangeably, clinical validation for each immune-therapy remains a necessity [9]. The interchangeability of PD-L1 assay must be considered carefully so as to ensure that no patient is devoid of treatment opportunity.

Inter-observer variability may lead to the obtainment of discordant results for PD-L1, which can consequentially impact therapy decisions. A recent study demonstrated that inter-observer agreement for each assay is moderate to high for IC staining (0.532–0.729) as well as TC staining (0.609–0.883) based on intra-class correlation coefficient obtained for UC [20]. However, 22C3 and 28-8 exhibited low inter-observer correlation in IC staining, while SP142 showed low inter-observer correlation in TC staining [20]. Downes et al. [21] suggested that excellent inter-observer agreement could be found using SP263 and 22C3, whereas PD-L1 scoring using SP142 was associated with a higher level of subjectivity in head and neck squamous cell carcinoma, breast carcinoma, and UC. The study of inter-observer heterogeneity of PD-L1 assays has also been well conducted using NSCLC samples. According to Cooper et al. [22], 10 pathologists reported good reproducibility at both 1% cutoffs of 22C3, whereas agreement was slightly lower for the 50% cutoff. Moreover, the Cardiopulmonary Pathology Study Group of the Korean Society of Pathologists investigated the inter-observer heterogeneity of PD-L1 staining with 22C3 using NSCLC samples [23]. Inter-observer reproducibility for the 1% cutoff was found to be relatively lower than the 50% cutoff, in contrast to the results reported by Cooper et al. [22]. Similarly, Rimm et al. [5] indicated that 13 pathologists reported excellent concordance when scoring TCs stained with any antibody (22C3, 28-8, SP142, and E1L3N) but reported poor concordance when scoring ICs stained with any antibody using NSCLC samples [16]. Although differing results were reported in previous studies of both UC and NSCLC, inter-observer heterogeneity of PD-L1 seems to occur.

UC is among the most histologically diverse cancers. A previous study found that infiltrating UC exhibits significantly higher T cell infiltration and PD-L1 expression than non-invasive papillary UC and UC in situ [24]. In addition to the conventional morphology observed, UC can contain elements of squamous differentiation, glandular differentiation, nested, plasmacytoid, sarcomatoid, and/or rarer variants. UC patients with histologic variants account for up to one-third of advanced cases. Li et al. [24] found that PD-L1 was expressed in a significant percentage of histologic variant of UC cases (cutoff 1% TC, 37% to 54%; cutoff 5% TC, 23% to 37%), while the highest PD-L1 expression was observed in patients with UC exhibiting squamous differentiation [25]. These results suggest that patients with histologic variants of UC may benefit more from anti–PD-1/PD-L1 therapy.

PD-1/PD-L1 target therapy has garnered considerable attention as a potential treatment strategy for patients with advanced UC. These agents are approved by the U.S. Food and Drug Administration for the treatment of patients with locally advanced or metastatic UC, with disease progression during or following platinum-containing chemotherapy, or disease progression within 12 months of neoadjuvant or adjuvant treatment with platinum-containing chemotherapy. Moreover, pembrolizumab and atezolizumab have received approval for first-line treatment of locally advanced or metastatic UC in patients ineligible for cisplatin-containing chemotherapy. PD-L1 expression levels in UC can thus effectively aid physicians in identifying patients who are more likely to benefit from anti–PD-1/PD-L1 therapy.

However, every single agent is tested in conjunction with a specific PD-L1 assay, which must be performed on a specific staining platform. Moreover, the PD-L1 scoring algorithm is heterogeneous and unique for each assay. These aspects have encouraged pathologists to consider assay interchangeability. Several studies have attempted to harmonize PD-L1 assays conducted for NSCLC and UC samples. However, SP142 was an outlier that detected markedly less PD-L1 expression in TC (Fig. 2A–C) [5–7,9,18,20]. In contrast, relatively high concordance was observed between SP263 and 22C3. As each assay is performed using different immunogens, and thus a unique epitope, different PD-L1 conformations or isoforms may lead to the obtainment of heterogeneous results. Moreover, the location of the antibody-binding domain has been known to affect the staining pattern, resulting in increased variability [26]. This discordance in the results suggests that the prospects for interchangeability of the assays is not optimistic. Although the discordance rate between SP263 and 22C3 is low, there is no scientific evidence to prove that two assays can be used interchangeably. It is of utmost priority that all patients should receive proper treatment, without any exceptions arising due to assay discordance.

The scoring algorithm for each PD-L1 assay in UC is described in Table 1 and Fig. 1. There are several common practical difficulties in analyses performed using PD-L1 assays. First, it is difficult to distinguish between TC and IC positivity. Representatively, distinguishing between TCs and macrophages proves burdensome when PD-L1 is stained, because macrophages are of comparative size to the TCs (Fig. 2H). Reviewing the hematoxylin and eosin-stained slides can be helpful in this respect. 22C3 should be used to count both ICs and TCs, which may not be problematic; however, SP142 should be excluded PD-L1–positive TCs. Moreover, ICs commonly include lymphocytes and macrophages. However, the ICs of SP263 additionally include plasma cells and neutrophils. Although this can also prove to be beneficial if we compare the hematoxylin and eosin-stained slides with PD-L1–stained slides, it remains uncertain whether PD-L1–positive neutrophils, plasma cells, and lymphocytes can be distinguished effectively (Fig. 2I). Lastly, the lamina propria at the base of the papillary lesion may contain lymphoid aggregates that show PD-L1 positivity, whereas only the lamina propria contiguous to the base of the tumor is considered part of the tumor area. Moreover, in fragmented tissue samples, including transurethral resection or biopsies—where distinction of intra- or peritumoral stroma cannot be ascertained—only stroma that is contiguous to individual tumor nests is included in the tumor area definition (Fig. 2G). However, the meaning of ‘contiguous’ is ambiguous and subjective.

The 22C3 scoring system uses the CPS algorithm. Theoretically, regardless of whether the number of TCs is large or small, 22C3 should be used to count all TCs present in the tumor area for the denominator. Additionally, all PD-L1–positive TCs and ICs should be counted in the tumor area for obtaining values for the numerator. This scoring algorithm is labor-intensive and an accurate calculation is practically impossible for the whole tumor area. Thus, the 22C3 manufacturer (Agilent) suggests that a partial portion of the tumor can be selected and scored. However, these results may be inconsistent due to intratumoral heterogeneity of 22C3 expression.

A comparatively higher positive cutoff (≥ 25% of TC) is required for SP263. If the test results do not meet the TC cutoff, the IC cutoff (≥ 25% of IC) can be evaluated subsequently. Unlike SP142, the total tumor area is not evaluated for IC scoring. IC positivity is only evaluated in the ICP. Therefore, SP263 results may easily meet the cutoff (≥ 25% of IC) because ICP is relatively smaller than the total tumor area values used as a denominator. However, the evaluation of complex geographic ICP may pose challenges. For accurate evaluation, one must physically draw and cut out the ICP—which lies beyond the confines of plausibility. Hence, advances in artificial intelligence and digital pathology are necessary for the precise assessment of SP263 expression. At present, SP263 is not clinically used as a companion diagnostic for durvalumab in UC.

SP142 evaluation for the determination of the IC-positive area and score seems relatively simpler than performing other assays. However, exclusion of SP142 TC positivity information from scoring may not be an appropriate approach. We analyzed three UC biopsy cases (unpublished data), which showed only strong PD-L1 positivity in TCs and not in ICs, with negative SP142 results (Fig. 2D–F). Although all the UC cases studied herein were SP142-negative, these patients demonstrated a complete response to atezolizumab. Whilst these results were obtained for only a limited number of cases, it should nonetheless be considered whether the exclusion of positive TCs from the scoring algorithm is a reasonable methodology. The ability of SP142 to detect TC expression is low, which may lead to the generation of statistical bias in clinical trials of atezolizumab in UC. This is probably main reason that TC were not included in the scoring algorithm of the SP142. Re-evaluation of SP142 TC expression may thus be necessary in a novel clinical study to assess anti–PD-1/PD-L1 therapy in UC. Furthermore, in cases where SP142 expression is observed only in TCs, it is recommended to adopt the 22C3 test to determine the applicability of pembrolizumab as first-line treatment.

These complex score algorithms and intratumoral heterogeneity of PD-L1 expression can result in inter-observer heterogeneity, particularly in scoring the SP142 of UC [20,21]. It is not difficult to score definite PD-L1 positive or negative cases; however, inter-observer heterogeneity must be observed in cases with approximate cutoff scores (e.g., 22C3 CPS, 5%–15%; SP142, 3%–10%, and so on). Providing training to the pathologist in these aspects as well as in the use of artificial intelligence may be a possible method for reduction of the inter-observer PD-L1 discordance.

Several studies have suggested the prognostic significance of PD-L1 expression in various malignancies; PD-L1–expressing tumors tend to exhibit poor prognosis [27]. Previous meta-analyses have demonstrated that PD-L1 expression is correlated with worse prognosis and advanced clinicopathological features in UC [28,29]. Moreover, Kawahara et al. indicated that UC with high-grade features exhibited higher PD-L1 expression [30]. A recent study has reported that the increased expression of PD-L1 is correlated with histologic variants of UC, including squamous, glandular, plasmacytoid, and sarcomatoid differentiation [25]. Histologic variants of UC constitute a high-grade feature that tends to be associated with PD-L1 expression. In this regard, PD-1/PD-L1 target therapy may be more effective for the treatment of histologic variants of UC. Although PD-L1 positivity is an important predictor of anti–PD-1/PD-L1 treatment response, PD-L1 expression status alone is insufficient to determine prognosis in any cancer subtype.

We reviewed the prospect and existing limitations of PD-L1 assays performed using samples from patients with UC. Discordance of PD-L1 positivity was observed, depending on the results of each assay. Notably, the inter-assay and inter-observer discordance were primarily observed in scoring SP142. Thus far, there exists no scientific evidence for the interchangeability of PD-L1 assays. If atezolizumab cannot be used as first-line therapy due to SP142 positivity observed only in TCs, and not in ICs, we recommend adopting the use of 22C3 in conjunction with pembrolizumab. The complex scoring algorithm of each assay is challenging for pathologists and also results in inter-observer heterogeneity. Providing suitable training to pathologists may be the only approach to overcome these challenges. Moreover, in the future, digital pathology and artificial intelligence may assist PD-L1 evaluation with greater accuracy.