Quantifying Intratumoral Biomarker Heterogeneity in Tubo-ovarian High-grade Serous Carcinoma to Optimize Clinical Translation

doi:10.21203/rs.3.rs-4726734/v1

Download PDF

Article

Quantifying Intratumoral Biomarker Heterogeneity in Tubo-ovarian High-grade Serous Carcinoma to Optimize Clinical Translation

https://doi.org/10.21203/rs.3.rs-4726734/v1

This work is licensed under a CC BY 4.0 License

You are reading this latest preprint version

Intratumoral heterogeneity (ITH) is spatial, phenotypic, or molecular differences within the same tumor that have important implications for accurate tumor classification and assessment of predictive biomarkers. The Canadian Ovarian Experimental Unified Resource (COEUR) has created a cohort of 437 FFPE tissue specimens from 108 tubo-ovarian high-grade serous carcinoma (HGSC) patients to quantify ITH across the anatomical sites and between primary and recurrence. We quantified the ITH of six clinically used immunohistochemical diagnostic and prognostic biomarkers (WT1, p53, p16, PR, CD8, and Ki67). Markers were stained on tissue microarrays and scored using a continuous or categorical interpretation of staining patterns. Two-way random effect and nested intraclass correlation were used to assess continuous markers, and Gwet’s AC1 was used for categorical markers. All biomarkers showed at least substantial agreement over several spatial comparisons, with WT1, p53 and p16 showing almost perfect agreement for most spatial comparisons. Similarly, categorical WT1, p53 and p16 showed almost perfect agreement for temporal comparisons, while the agreement for primary versus recurrence for PR, CD8 and Ki67 was only fair. We provide power calculations to achieve reliability of >0.60 and recommend testing emerging protein biomarkers to see whether they reach a clinically acceptable benchmark level of ITH.

Biological sciences/Cancer/Tumour heterogeneity

Health sciences/Oncology/Cancer/Tumour biomarkers

Health sciences/Oncology/Cancer/Gynaecological cancer/Ovarian cancer

Ovarian cancer

high-grade serous

intratumoral heterogeneity

TP53

CD8

WT1

Over the past decade, due to advancements in various omics and imaging technologies, there has been a surge in the discovery of cancer biomarkers. However, the clinical adoption of these biomarkers has been limited by reproducibility challenges[1]. One of the significant factors affecting reproducibility is intra-tumor heterogeneity (ITH), which is within-patient biomarker differences, in contrast to inter-tumor heterogeneity, which measures variability across patients.

ITH is a complex phenomenon that can arise from various biological and technical factors and can be observed in cellular morphology, molecular alterations, and epigenetic variability.

Biological factors impacting ITH include genomic instability, clonal evolution, and anatomical site-specific differences in the microenvironment[2]. Genomic instability is a hallmark of tubo-ovarian high-grade serous carcinoma (HGSC), which can result in the accumulation of genetic mutations and copy number changes in different regions of the tumor, leading to the emergence of varying tumor subclones with distinct genetic profiles [3–5]. For example, in 22% of HGSC patients, the homologues repair deficiency score would have been positive (cut-off ≥ 42) in some tumor samples and negative in others, changing the final result depending on which tumor site would have been evaluated[6] Over time, tumors can evolve and adapt to changing microenvironmental conditions, leading to new subclones that may express different biomarkers. Additionally, interactions between tumor and microenvironment, including immune cells and stromal cells, can influence the expression of biomarkers at different anatomical sites of metastatic disease. While accessing HGSC as an omental core biopsy is thought to be safer than puncturing a cystic ovarian tumor with the potential of spillage, the microenvironment of the two anatomical sites differs[7].

ITH can, therefore, have direct clinical implications for cancer diagnosis, prognosis, treatment selection and the understanding of treatment resistance. Moreover, ITH impacts the power of studies and can attenuate markers' prognostic or predictive value of biomarkers. For diagnostic markers, ITH, if present, can result in inaccurate tumor classification that hinders clinical translation and minimizes the utility of biomarkers to guide patient management[8, 9]. Additionally, ITH serves as a mechanism of therapeutic resistance and treatment failure[10], and its presence has been consistently associated with unfavourable outcomes among cancer patients with metastatic disease[11] For biomarkers to effectively monitor and detect earlier biological changes or enhance treatment decision-making, they must be valid, reliable and practical. This highlights the importance of quantifying ITH, an often-overlooked step in biomarker validation when deriving prognostic signatures, determining risk scores, or designing effective treatment strategies.

This paper introduces HGSC tissue microarrays (TMA) cohorts from the Comprehensive Ovarian Cancer Research (COEUR) designed to model in-situ, inter-anatomical site, and temporal ITH of immunohistochemistry (IHC) biomarkers. COEUR is a pan-Canadian initiative launched by the Terry Fox Research Institute (TFRI) in 2015 to improve outcomes for women with ovarian cancer by accelerating the translation of research discoveries into clinical practice [12]. We selected four IHC diagnostic biomarkers (WT1, p53, p16, PR)[13, 14] and two prognostic markers (CD8 and Ki67) [15, 16] to evaluate ITH in these established clinical biomarkers, which can serve as benchmarks for novel biomarker discoveries[17–21]

Our investigation encompassed 430 specimens from 106 unique high-grade serous tubo-ovarian carcinoma patients. Three TMAs were prepared for different heterogeneity analyses (Fig. 1 and Supplementary Table S1). The first TMA was designed to investigate in-situ heterogeneity, referring to variability within primary tubo-ovarian tumor sites (Supplementary Fig. S1); it consisted of primary chemo-naïve tubo-ovarian tumor FFPE blocks from 21 patients[22] (3 blocks per patient). The second TMA included samples from 59 unique patients and was designed to study anatomical site heterogeneity between adnexal sites, including the fallopian tube and ovary, and the omentum (Fig. 2) or other peritoneal sites (Supplementary Fig. S2). A third (from COEUR) and fourth (from the University of Calgary) TMA were used to analyze temporal heterogeneity in 14 patients each, where data was compared between paired specimens from a primary debulking surgery (chemo-naïve specimen from adnexa or omentum, etc.) to a secondary debulking surgery of a recurrence from a metastasis specimen including but not limited to peritoneal sites (Supplementary Fig. S3). Table 1 summarizes the prevalence of each biomarker under different conditions (different anatomical sites, different time points, etc.).

Table 1

**Biomarker Prevalence by Anatomical Site**. This table details the prevalence of biomarkers across various anatomical sites. For each categorical biomarker, prevalence is represented as a proportion or percentage. The table includes quantitative biomarkers' mean expression level and corresponding standard deviation.
	In-Situ	Anatomical Site			Temporal
Biomarker	≥ 3 Blocks n = 48	Adnexa n = 74	Omentum n = 62	Other peritoneal n = 33	Primary n = 27	Recurrence n = 27
WT1 (%)	87.4 (25.1)	92.2 (18.6)	90.6 (24.1)	94.7 (15.9)	78.9 (30.6)	80.0 (28.0)
Unknown		1	2	1
WT1 (cat.)
Absence	2 (4.2%)	0 (0%)	2 (3.3%)	0 (0%)	1 (3.7%)	2 (7.4%)
Presence	46 (96%)	73 (100%)	58 (97%)	32 (100%)	26 (96%)	25 (93%)
Unknown		1	2	1
p53 (cat.)
Normal wild type pattern	1 (2.1%)	0 (0%)	0 (0%)	0 (0%)	0 (0%)	0 (0%)
Abnormal	47 (98%)	66 (100%)	55 (100%)	29 (100%)	27 (100%)	27 (100%)
Unknown		8	7	4
p16 (cat.)
Abnormal complete absence	6 (13%)	4 (6.1%)	3 (5.4%)	1 (3.6%)	2 (7.4%)	2 (7.4%)
Normal patchy	7 (15%)	22 (33%)	14 (25%)	9 (32%)	8 (30%)	6 (22%)
Abnormal block	35 (73%)	40 (61%)	39 (70%)	18 (64%)	17 (63%)	19 (70%)
Unknown		8	6	5
PR (%)	6.9 (12.3)	5.9 (11.5)	2.6 (7.7)	1.0 (3.0)	8.2 (17.3)	3.6 (8.9)
Unknown		0	1	1
PR (cat.)
Absence	27 (56%)	44 (59%)	45 (74%)	27 (84%)	15 (56%)	16 (59%)
Presence	21 (44%)	30 (41%)	16 (26%)	5 (16%)	12 (44%)	11 (41%)
Unknown		0	1	1
CD8 (count per HPF)	7.7 (7.9)	11.4 (11.8)	11.9 (13.1)	9.0 (9.8)	11.1 (12.0)	19.1 (18.8)
Unknown		8	6	4
CD8 (categorical)
< median	24 (50%)	26 (39%)	26 (46%)	15 (52%)	12 (44%)	10 (37%)
≥ median	24 (50%)	40 (61%)	30 (54%)	14 (48%)	15 (56%)	17 (63%)
Unknown		8	6	4
Ki67 (%)	30.7 (17.5)	35.6 (18.9)	38.5 (21.3)	42.2 (18.8)	34.1 (23.7)	36.0 (24.0)
Unknown		9	7	4	1	0
Ki67 (cat.)
< median	27 (56%)	25 (38%)	22 (40%)	8 (28%)	13 (50%)	10 (37%)
≥ median	21 (44%)	40 (62%)	33 (60%)	21 (72%)	13 (50%)	17 (63%)
Unknown		9	7	4	1	0
¹ Mean (SD); n (%)

Evaluating ITH

Evaluating ITH requires assessing the variability of k repeated biomarker measurements for each patient under various conditions. High variability among these measurements indicates low agreement and high heterogeneity. To quantify the agreement of quantitative measurements from the same subject across multiple trials, we use the intraclass correlation coefficient (ICC)[23]. ICC is the ratio of between-subject variance to total variance and can also be interpreted as the agreement of measurements within-subject. ICC values range from 0 to 1, with higher values indicating good agreement and lower values suggesting the presence of heterogeneity. When measurements take on categorical values, reliability is typically measured with Cohen’s Kappa and its extensions (e.g., Fleiss’ Kappa and Light’s Kappa)[24]. However, recently, Gwet’s AC1[25] has emerged as an alternative measure, outperforming kappa measures because of its robustness against the "Kappa paradox"[26, 27], which occurs in the presence of class imbalance, where one of the measurement categories exhibits higher prevalence than the others. This can result in low kappa, even when the agreement is high. This phenomenon is also observed when measurements exhibit very little heterogeneity. Like ICC, Gwet’s AC1 takes on values between 0 and 1, with higher values indicating increased reliability. We can define ITH as 1 - reliability. Even though the interpretation of the reliability measure is ad-hoc[28], and the various measures do not have a one-to-one mapping, we define an interpretation in Table 2, which we adopt and use throughout.

Table 2

**Agreement and Heterogeneity Mapping**. This table illustrates the relationship between levels of agreement and heterogeneity. Descriptive labels for levels of agreement are provided to aid in interpretation. Corresponding levels of heterogeneity are shown in a mirrored format for comparison.
Reliability Value	Agreement	Heterogeneity
-1.0–0.0	Poor	Almost complete
0.01–0.20	Slight	Almost complete
0.21–0.40	Fair	Substantial
0.41–0.60	Moderate	Moderate
0.61–0.80	Substantial	Fair
0.81–1.00	Almost perfect	Slight

Table 3 reports the two-way mixed effects ICC for continuous markers and Gwet’s AC1 score for categorical markers. Supplementary Table S2 presents Fleiss’ Kappa results for comparison; Supplementary Table S3 presents ICC results computed from nested mixed effects models.

Table 3

Reliability Summary by Biomarker
Biomarker	ITH Type	# patients Complete/Assessed	Reliability Model	Estimate (95% CI)	Interpretation
Minimal Diagnostic Biomarker
WT1	In-situ	47/48	ICC(A,1)	0.75 (0.63,0.84)	Substantial
	Adnexa vs. Omentum	60/62		0.67 (0.51,0.79)	Substantial
	Adnexa vs. Other peritoneal	31/33		0.72 (0.49,0.85)	Substantial
	Primary vs. Recurrence	27/27		0.79 (0.59,0.9)	Substantial
WT1 (cat.)	In-situ	47/48	AC1	0.92 (0.848,0.996)	Almost Perfect
	Adnexa vs. Omentum	60/62		0.95 (0.881,1)	Almost Perfect
	Adnexa vs. Other peritoneal	31/33		1 (1,1)	Almost Perfect
	Primary vs. Recurrence	27/27		1 (1,1)	Almost Perfect
p53	In-situ	45/48	AC1	1 (1,1)	Almost Perfect
	Adnexa vs. Omentum	55/62		1 (1,1)	Almost Perfect
	Adnexa vs. Other peritoneal	29/33		1 (1,1)	Almost Perfect
	Primary vs. Recurrence	27/27		1 (1,1)	Almost Perfect
p16	In-situ	46/48	AC1	0.86 (0.76,0.96)	Almost Perfect
	Adnexa vs. Omentum	56/62		0.88 (0.78,0.99)	Almost Perfect
	Adnexa vs. Other peritoneal	28/33		0.68 (0.44,0.92)	Substantial
	Primary vs. Recurrence	27/27		0.9 (0.77,1)	Almost Perfect
PR	In-situ	43/48	ICC(A,1)	0.76 (0.64,0.85)	Substantial
	Adnexa vs. Omentum	61/62		0.7 (0.5,0.83)	Substantial
	Adnexa vs. Other peritoneal	32/33		0.12 (-0.2,0.43)	Slight
	Primary vs. Recurrence	27/27		0.24 (-0.13,0.55)	Fair
PR (cat.)	In-situ	43/48	AC1	0.51 (0.31,0.715)	Moderate
	Adnexa vs. Omentum	61/62		0.58 (0.369,0.798)	Moderate
	Adnexa vs. Other peritoneal	32/33		0.76 (0.526,0.987)	Substantial
	Primary vs. Recurrence	27/27		0.33 (-0.073,0.74)	Fair
Potentially Prognostic Biomarker
CD8	In-situ	47/48	ICC(A,1)	0.58 (0.42,0.72)	Moderate
	Adnexa vs. Omentum	56/62		0.56 (0.35,0.71)	Moderate
	Adnexa vs. Other peritoneal	29/33		0.7 (0.46,0.85)	Substantial
	Primary vs. Recurrence	27/27		0.45 (0.1,0.7)	Moderate
CD8 (cat.)	In-situ	47/48	AC1	0.49 (0.301,0.686)	Moderate
	Adnexa vs. Omentum	56/62		0.46 (0.225,0.704)	Moderate
	Adnexa vs. Other peritoneal	29/33		0.66 (0.365,0.949)	Substantial
	Primary vs. Recurrence	27/27		0.26 (-0.13,0.65)	Fair
Ki67	In-situ	46/48	ICC(A,1)	0.61 (0.46,0.75)	Substantial
	Adnexa vs. Omentum	54/62		0.63 (0.44,0.77)	Substantial
	Adnexa vs. Other peritoneal	29/33		0.5 (0.18,0.73)	Moderate
	Primary vs. Recurrence	26/27		0.34 (-0.05,0.64)	Fair
Ki67 (cat.)	In-situ	46/48	AC1	0.37 (0.167,0.579)	Fair
	Adnexa vs. Omentum	54/62		0.41 (0.162,0.667)	Moderate
	Adnexa vs. Other peritoneal	29/33		0.45 (0.082,0.809)	Moderate
	Primary vs. Recurrence	26/27		0.24 (-0.168,0.638)	Fair

Diagnostic Biomarkers

The first-line, confirmatory diagnostic biomarkers for HGSC, WT1 and p53, showed very strong in situ agreement and minimal ITH (Table 3). Agreement among categorical measures of WT1 was almost perfect despite some variation (substantial agreement) for continuous assessment of WT1. The categorical assessment of WT1 using Fleiss’ kappa was very weak despite the considerable agreement observed in the data, highlighting the limitation of kappa in the presence of little variability (Supplementary Table S2). Gwet’s AC1 scores, in comparison, corresponded better with the agreement observed in the data and from the continuous assessment. Similarly, categorical p53 results were consistent across all types of ITH, exhibiting perfect agreement. We also saw little heterogeneity for the second-line diagnostic biomarker p16; agreement for most types of ITH was still almost perfect, except for the comparison of adnexa to other peritoneal, where the ICC was slightly lower at 0.68 (95% CI: 0.44–0.92, Table 2).

The second-line diagnostic and prognostic marker PR showed the highest variability across all biomarkers. While agreement between anatomical sites varied from slight to moderate and substantial (Table 3), the temporal agreement for PR was only fair but with a similar prevalence at primary (44%) or recurrence (41%) (Table 1).

Similar results were observed with ICC computed from nested mixed effects models (Supplementary Table S3), even though some results failed to converge due to numerical issues resulting from the limited heterogeneity observed.

Prognostic Biomarkers

Continuous prognostic marker CD8 had the highest reliability ranging from moderate to substantial (ICC = 0.7 (95%CI: 0.46–0.85)) for adnexa vs. other peritoneal disease, whereas continuous Ki67 exhibited the highest reliability both in situ (ICC = 0.61 (95% CI: 0.46–0.75)) and for adnexa vs. omentum (ICC = 0.63 (95% CI: 0.44–0.77), Table 3). Categorical versions of CD8 and Ki67, where the biomarkers were dichotomized at the median, showed lower reliability across all comparisons. The categorical temporal agreement for CD8 and Ki67 was only fair with increased CD8 counts per HPF from 11.1 to 19.1 but without showing a significant change in the Ki67 labelling index from primary (34%) to recurrent (36%) (Table 1).

Within the nested mixed effects models in Table S4, the ICC for both continuous and categorical measures of CD8 and Ki67 was generally lower compared with diagnostic biomarkers. However, these reliability scores were higher than ICC and Gwet’s AC1 score computed from the two-way mixed effects framework across all heterogeneity comparisons.

A summary of the reliability of all markers is illustrated in Fig. 3 in decreasing order of ITH.

Designing validation studies for ITH

For a biomarker to not exhibit ITH, minimal agreement must be achieved between the repeated measurements. Below, we detail the lowest level of agreement that each TMA is powered to estimate. Using a value of κ₀ = 0.8 under the null hypothesis of no heterogeneity, we estimated the lower bound of detectable κ_L for binary, 3-level, and 4-level outcomes, each with three possible prevalence patterns (Table 4). We use the number of samples and raters (conditions) observed in the data for estimating κ_L. Prevalence is one of the primary drivers of statistical power, with biomarkers expressed in a smaller proportion of the population (low prevalence) requiring a larger sample size and more measurements to achieve sufficient power to detect more significant levels of agreement.

Table 4

A lower bound of agreement that can be computed for different types of ITH under various prevalence scenarios.
ITH	N	k	Binary			3 categories			4 categories
ITH	N	k	0.05	0.2	0.4	A¹	B²	C³	D⁴	E⁵	F⁶
In situ	19	4	0.25	0.52	0.59	0.63	0.56	0.63	0.65	0.60	0.64
In situ	48	3	0.41	0.61	0.66	0.69	0.63	0.68	0.69	0.66	0.69
In situ	76	2	0.43	0.62	0.65	0.68	0.63	0.68	0.69	0.66	0.69
Adnexa vs. Omentum	62	2	0.39	0.59	0.63	0.67	0.60	0.67	0.68	0.64	0.67
Adnexa vs. Other peritoneal	33	2	0.26	0.49	0.56	0.61	0.52	0.61	0.63	0.57	0.62
Primary vs. Recurrence	27	2	0.22	0.46	0.52	0.59	0.48	0.58	0.61	0.54	0.60
¹ p=[1/3, 1/3, 1/3]. ² p=[0.8, 0.1, 0.1]. ³ p=[0.4, 0.4, 0.2]. ⁴ p=[1/4, 1/4, 1/4, 1/4]. ⁵ p=[0.7, 0.1, 0.1, 0.1]. ⁶ p=[0.4, 0.3, 0.2, 0.1].

We have established a TMA resource to assess the in situ, inter-anatomic site, and temporal ITH of IHC biomarkers in HGSC. We observed variations of several biomarkers across different anatomical sites and temporal comparisons between primary and recurrence. Using clinically established IHC markers, we provided a benchmark for ITH that can be used to investigate new emerging biomarkers.

WT1 and p53 represent the minimal ancillary IHC panel to confirm a diagnosis of HGSC [13]. As expected, for the truncal founder mutation of HGSC [29], p53 showed extremely low ITH across anatomical sites and temporally distinct samples. A similar high agreement was previously demonstrated for pre- and post-chemotherapy HGSC samples [21]. Rarely, two different abnormal p53 staining patterns can be observed in what seems a single tumor, but this should raise suspicion for the dual primary origin of two HGSC from different STICs or co-occurrence of HGSC with an endometrial serous carcinoma [30] Subclonal p53 IHC, defined by the combination of abnormal with normal wild-type pattern, is inconsistent with a diagnosis of HGSC and should raise the possibility of an alternative diagnosis, such as endometrioid carcinoma. However, this must be distinguished from areas of poor fixation within HGSC [31]

Similarly, WT1 has also been shown to be robustly expressed in pre- and post-chemotherapy HGSC samples[21] and herein, we show almost perfect agreement using categorical data. However, in a continuous analysis, the agreement dropped to only moderate for some types of heterogeneity. Although pathologists consider the extent of expression in their interpretation (i.e., integrated with the morphological appearance of a tumor), this should only cause occasional practical problems since a categorized interpretation (absent versus present) remains the standard in clinical pathology reporting. Most HGSCs show diffuse WT1 in all tumor cells. Yet in a few HGSC samples, a subset of the tumor cells may lose WT1 expression (WT1 is expressed in a normal fallopian tube, the tissue of origin, in both secretory and ciliated cells), potentially due to the effects of chromosomal instability/copy number variations, including homozygous deletions of WT1[32]. While WT1 shows slight ITH as a continuous marker, it is very robust as a categorical marker.

The concordance for p16, sometimes used as a second-line diagnostic marker[14], was also almost perfect and showed only slightly lower agreement in comparison with p53. Abnormal IHC expression for p16 in HGSC is multicausal and includes homozygous deletion for the abnormal complete absence pattern, while abnormal block expression may be caused by alterations in the G1/S phase such as RB1 loss [33] and CCNE1 high-level amplification[34]. RB1 loss, for example, can occasionally occur in a subclonal fashion, which could result in ITH of p16[33]

The potentially prognostic markers showed a weaker categorical concordance, especially in the temporal context, when comparing primary and recurrence, potentially due to systematic effects of in-between treatment or oncogenic changes during tumor evolution and progression. For example, neoadjuvant chemotherapy lowers the Ki67 labelling index of HGSC[35] The primary samples from the COEUR cohort came exclusively from chemo-naïve primary debulking specimens[36] While we observed significant ITH when comparing primary and recurrence, we did not observe systematic shifts in the prevalence for the Ki67 labelling index on recurrence. This suggests that in contrast to the short time interval between neoadjuvant chemotherapy and interval debulking, proliferation rates recover in case of recurrence, which usually occurs sometime after chemotherapy exposure. There was also no change in the prevalence of PR in the temporal course, but CD8 counts increased on recurrence. While a favourable prognostic marker should not necessarily increase on recurrence[16] an increase of CD3 and CD4 lymphocytes in recurrent HGSC has been reported in another study[37] This poorer temporal agreement among the prognostic markers may require retesting on recurrence with the need to obtain additional longitudinal tissue biopsies.

Within the spatial context, however, the potential prognostic biomarker (PR, CD8, Ki67) generally showed a substantial agreement between anatomical sites, which supports using a single biopsy for adequate biomarker assessment. For CD8, this aligns with another study that concluded that small tissue biopsies represent the immune microenvironment adequately [38] Overall, all six tested IHC biomarkers showed substantial agreement with ICC > 0.60 in most spatial comparisons. This compares favourably to the recent ITH in HRD scores, with only 78% agreement [6] However, as in this example, results on ITH are often not comparable due to differences in reporting (% agreement, kappa, ICC). We recommend reporting ICC as the best way to report ITH, with values over 0.60 as being clinically robust. We provide a power calculation to estimate the number of samples needed to assess for ITH.

In addition to true biological factors, technical factors can contribute to ITH, such as the quantity and quality of tumor sampling. Tissue fixation and processing can impact biomarker expression levels, as non-standardized fixation protocols and processing methods may affect the stability and accessibility of biomolecules. Differences in the quantity of tumor cells available for analysis and tumor cellularity can also result in heterogeneity.

In conclusion, molecular analysis for diagnostic, prognostic or predictive biomarkers is typically conducted on one random tumor sample extracted from patients with HGSC. This method is valid if molecular alterations are truncal and present in every tumor cell. However, several driver alterations occur during tumor evolution in subclones [3], and their accurate assessment will be subject to sampling bias. While this bias is unavoidable, it should be at least measured, and if the acceptable ICC threshold of > 0.60 cannot be met, mitigating strategies such as increase in number of samples tested, minimum sample size, standardization of the sample site and time (primary, possible retesting on recurrence) should be established.

Study Population

Formalin-fixed paraffin-embedded (FFPE) tumor tissue specimens from the TFRI COEUR and a cohort from the University of Calgary were obtained from 108 patients diagnosed with HGSC. The COEUR national ovarian cancer database collects clinical, pathological, and genomic data from patients with ovarian cancer. COEUR uses biobank material; each biobank received ethics approval from local review boards to collect and share samples and clinical data. All subjects gave broad written informed consent to future research with their samples and data without restriction. Additionally, the central activities of the Centre de recherche du Centre Hospitalier de l’Université de Montréal, which include the collection of the COEUR repository samples and data, received local ethics approval from the ‘Comité d’éthique de la recherche du CHUM (project reference: 2010-3552-9H (80306)). We also accessed a TMA with tissue from 14 HGSC participants obtained from primary surgery and recurrence from the University of Calgary. Ethics approval and a waiver of informed consent were received from the Health Research Ethics Board of the Alberta Cancer Committee (HREBA.CC-16-0371). All methods were carried out in accordance with relevant guidelines and regulations.

TMA preparation and study design

Three core punches 0.6 mm were obtained from each block. A detailed description of specimens on each TMA is described in Supplementary Table S1. The IHC protocols are provided in Supplementary Table S2. Two gynecologic subspecialty pathologists (RK for COEUR, MK for University of Calgary samples) reviewed and confirmed the HGSC diagnoses before TMA construction and IHC staining. Only cases with at least 30% tumor cellularity in the area of interest were included. MK scored all IHC markers. Counts of CD8 intraepithelial lymphocytes per 400x high power field (HPF) were treated as continuous but were capped at 50 to mitigate outliers. WT1, Ki67, and PR expression levels were visually scored in 5% intervals with finer intervals at extremes. p16 expression was categorized as "Abnormal complete absence”, “Normal patchy”, or "Abnormal Block" [39], while p53 was dichotomized into "Abnormal" (overexpression, complete absence, or cytoplasmic pattern) and "Normal wild type pattern” [14, 29] We also presented a categorical version of the continuous biomarkers, with CD8 and Ki67 dichotomized at the overall median (≥ 5/HPF for CD8) and (≥ 30% for Ki67). WT1 and PR were dichotomized as absent or present (≥ 1%).

A priori and post-hoc power calculation

A sample size calculation was conducted before TMA construction to inform their design. We adopted a conservative approach by assuming binary biomarkers. We fixed the Type 1 error rate at 0.05. We computed the number of subjects needed to achieve 80% power for prevalences ranging from 0.05 to 0.5 and different numbers of specimens ranging from 2 to 6, using the R project package (R-4.3.3)[40] kappaSize (v1.2) (Supplementary Table S3). Under the null hypothesis, biomarkers were assumed to have a strong agreement and minimal heterogeneity (agreement of 0.8 or higher). Reliability values below 0.6 are considered inadequate, leading to high heterogeneity and low confidence[41] we used 0.6 as the lower limit for the one-sided 95% confidence interval.

As part of this project, we also provided, for future studies, the minimal lower bound for an intra-class kappa that can be detected when using these TMAs to evaluate ITH in categorical biomarkers with 2, 3 and 4 categories, expressed with different prevalence (Table 4).

Statistical analysis

We evaluated ITH using ICC for continuous biomarkers, and for categorical biomarkers, we used Fleiss’ Kappa (which generalizes Cohen’s Kappa to more than two categories) and Gwet’s AC1. To evaluate in situ heterogeneity, we included patients with at least three blocks from adnexal tumors. For patients with more than three blocks available, we randomly selected three blocks. The k repeated core measurements within a tumor block were collapsed into a single score by taking the average value for continuous markers or the mode (most frequent value) for categorical markers. When a tie was observed, we randomly selected one of the values. A sensitivity analysis was performed by repeating the analysis and choosing the alternative most frequent value to verify that this random selection did not significantly impact the results. Patients with fewer than three blocks were excluded from the analysis.

For anatomical site heterogeneity, we compared the adnexa tumor to the omentum and the adnexa tumor to the other peritoneal site. Similarly, to evaluate temporal heterogeneity, we compared biomarkers from primary and recurrence tumors. When considering markers from different conditions (anatomical site or temporal heterogeneity), we collapsed scores from different blocks within each anatomical site or time point. To be included in this analysis, patients had to have at least one block for each condition considered. We computed ICC using 2-way mixed effects with an absolute agreement and followed the recommendations for repeated test-retest measurements in reliability analysis [42]. Calculations used the IRR (v0.84.1) package in R[43]. We computed Fleiss’ kappa and AC1 for categorical biomarkers using the irrCAC (v1.0) package [44].

Averaging measurements within a block or condition may result in a loss of information. To avoid average and account for the nested structure of the collected measurements, we also computed ICC using mixed effects models with R packages lme4 (v1.1-35.3) and the corresponding confidence intervals using performance (v0.12.0). For in-situ heterogeneity, we fit a mixed model with a fixed effect for block and nesting biomarker measurements from the same patient within distinct blocks. We used linear mixed effects models for continuous biomarkers and binomial generalized linear mixed effects models for categorical biomarkers. For anatomical site heterogeneity, we fit a mixed effects model with a fixed effect for the anatomical site and nesting scores from the same patient within distinct blocks and the blocks, in turn, are nested in different anatomical sites. For temporal heterogeneity, we fit a mixed effects model with a fixed effect for time and nesting scores from the same patient within the same time point. We partitioned variance components arising from different sources of uncertainty[45] We computed the adjusted ICC as the proportion of variance arising from random effects when considering a total variance that includes variance from random effects and residuals. Models that failed due to singularity issues due to low variability could not be assessed for nested ICC. We reported model-based bootstrap confidence intervals.

Additional Information

Dr. Kobel is a HelixBiopharma consultant. Dr. Talhouk is a Health Research BC Scholar. All co-authors declare no conflict of interest.

Author Contribution

AT, DP, DGH, AMM, and MK conceived, designed and supervised the study. KL, RM, CLP, MB and MK identified HGSC cases, reviewed morphology and created the tissue microarray resource; AT and DC performed statistical analyses and drafted the manuscript. All authors revised the manuscript and approved its final version.

Acknowledgement

This study used resources supported by the Terry Fox Research Institute and Ovarian Cancer Canada and managed and supervised by the Centre Hospitalier de l’Universite de Montreal (CRCHUM). The TFRI COEUR cohort acknowledges the contributions from Institutions across Canada (for a full list, see http:// www.tfri.ca/en/research/translational-research/coeur/ coeur_biobanks.aspx).

Data Availability

The datasets generated and analyzed during the current study are not publicly available, due to privacy laws, but are available from the corresponding author upon reasonable request. The generated data and the TMAs can also be requested from TFRI COEUR ([email protected]), but restrictions apply to the availability of these data, which were used under license for the current study and so are not publicly available.

Boutros PC. The path to routine use of genomic biomarkers in the cancer clinic. Genome Res. 2015;25(10):1508–1513. doi:10.1101/GR.191114.115
Hunt AL, Bateman NW, Barakat W, et al. Extensive three-dimensional intratumor proteomic heterogeneity revealed by multiregion sampling in high-grade serous ovarian tumor specimens. iScience. 2021;24(7). doi:10.1016/J.ISCI.2021.102757
Khalique L, Ayhan A, Weale ME, Jacobs IJ, Ramus SJ, Gayther SA. Genetic intra-tumour heterogeneity in epithelial ovarian cancer and its implications for molecular diagnosis of tumours. J Pathol. 2007;211(3):286–295. doi:10.1002/PATH.2112
Bashashati A, Ha G, Tone A, et al. Distinct evolutionary trajectories of primary high-grade serous ovarian cancers revealed through spatial mutational profiling. J Pathol. 2013;231(1):21–34. doi:10.1002/PATH.4230
Schwarz RF, Ng CKY, Cooke SL, et al. Spatial and temporal heterogeneity in high-grade serous ovarian cancer: a phylogenetic analysis. PLoS Med. 2015;12(2). doi:10.1371/JOURNAL.PMED.1001789
Cunnea P, Curry EW, Christie EL, et al. Spatial and temporal intra-tumoral heterogeneity in advanced HGSOC: Implications for surgical and clinical outcomes. Cell Rep Med. 2023;4(6). doi:10.1016/J.XCRM.2023.101055
Talhouk A, George J, Wang C, et al. Development and Validation of the Gene Expression Predictor of High-grade Serous Ovarian Carcinoma Molecular SubTYPE (PrOTYPE). Clinical Cancer Research. 2020;26(20):5411–5423. doi:10.1158/1078-0432.CCR-20-0103
Dunne PD, McArt DG, Bradley CA, et al. Challenging the cancer molecular stratification dogma: Intratumoral heterogeneity undermines consensus molecular subtypes and potential diagnostic value in colorectal cancer. Clinical Cancer Research. 2016;22(16):4095–4104. doi:10.1158/1078-0432.CCR-16-0032/116033/AM/CHALLENGING-THE-CANCER-MOLECULAR-STRATIFICATION
Allott EH, Geradts J, Sun X, et al. Intratumoral heterogeneity as a source of discordance in breast cancer biomarker classification. Breast Cancer Res. 2016;18(1). doi:10.1186/S13058-016-0725-1
Walmsley CS, Jonsson P, Cheng ML, et al. Convergent evolution of BRCA2 reversion mutations under therapeutic pressure by PARP inhibition and platinum chemotherapy. npj Precision Oncology 2024 8:1. 2024;8(1):1–6. doi:10.1038/s41698-024-00526-9
Ramón y Cajal S, Sesé M, Capdevila C, et al. Clinical implications of intratumor heterogeneity: challenges and opportunities. Journal of Molecular Medicine 2020 98:2. 2020;98(2):161–177. doi:10.1007/S00109-020-01874-2
Le Page C, Rahimi K, Köbel M, et al. Characteristics and outcome of the COEUR Canadian validation cohort for ovarian cancer biomarkers. BMC Cancer. 2018;18(1). doi:10.1186/S12885-018-4242-8
Köbel M, Rahimi K, Rambau PF, et al. An Immunohistochemical Algorithm for Ovarian Carcinoma Typing. Int J Gynecol Pathol. 2016;35(5):430–441. doi:10.1097/PGP.0000000000000274
Altman AD, Nelson GS, Ghatage P, et al. The diagnostic utility of TP53 and CDKN2A to distinguish ovarian high-grade serous carcinoma from low-grade serous ovarian tumors. Mod Pathol. 2013;26(9):1255–1263. doi:10.1038/MODPATHOL.2013.55
Chen M, Yao S, Cao Q, Xia M, Liu J, He M. The prognostic value of Ki67 in ovarian high-grade serous carcinoma: an 11-year cohort study of Chinese patients. Oncotarget. 2016;8(64):107877–107885. doi:10.18632/ONCOTARGET.14112
Goode EL, Block MS, Kalli KR, et al. Dose-Response Association of CD8 + Tumor-Infiltrating Lymphocytes and Survival Time in High-Grade Serous Ovarian Cancer. JAMA Oncol. 2017;3(12). doi:10.1001/JAMAONCOL.2017.3290
Lara OD, Krishnan S, Wang Z, et al. Tumor core biopsies adequately represent immune microenvironment of high-grade serous carcinoma. Sci Rep. 2019;9(1). doi:10.1038/S41598-019-53872-1
Hagemann AR, Hagemann IS, Cadungog M, et al. Tissue-based immune monitoring II: multiple tumor sites reveal immunologic homogeneity in serous ovarian carcinoma. Cancer Biol Ther. 2011;12(4):367–377. doi:10.4161/CBT.12.4.16908
Köbel M, Turbin D, Kalloger SE, Gao D, Huntsman DG, Gilks CB. Biomarker expression in pelvic high-grade serous carcinoma: comparison of ovarian and omental sites. Int J Gynecol Pathol. 2011;30(4):366–371. doi:10.1097/PGP.0B013E31820D20BA
Bekos C, Pils D, Dekan S, et al. PD-1 and PD-L1 expression on TILs in peritoneal metastases compared to ovarian tumor tissues and its associations with clinical outcome. Sci Rep. 2021;11(1). doi:10.1038/S41598-021-85966-0
Casey L, Köbel M, Ganesan R, et al. A comparison of p53 and WT1 immunohistochemical expression patterns in tubo-ovarian high-grade serous carcinoma before and after neoadjuvant chemotherapy. Histopathology. 2017;71(5):736–742. doi:10.1111/HIS.13272
Brassard J, Hughes MR, Dean P, et al. A tumor-restricted glycoform of podocalyxin is a highly selective marker of immunologically cold high-grade serous ovarian carcinoma. Front Oncol. 2023;13. doi:10.3389/FONC.2023.1286754
McGraw KO, Wong SP. Forming Inferences about Some Intraclass Correlation Coefficients. Psychol Methods. 1996;1(1):30–46. doi:10.1037/1082-989X.1.1.30
Zhao X, Feng GC, Ao SH, Liu PL. Interrater reliability estimators tested against true interrater reliabilities. BMC Med Res Methodol. 2022;22(1). doi:10.1186/s12874-022-01707-5
Li Gwet K. HANDBOOK OF INTER-RATER RELIABILITY Fourth Edition The Definitive Guide to Measuring the Extent of Agreement Among Raters.
Zec S, Soriani N, Comoretto R, Baldi I. High Agreement and High Prevalence: The Paradox of Cohen’s Kappa. Open Nurs J. 2017;11(1):211–218. doi:10.2174/1874434601711010211
Xie Q. Agree or Disagree? A Demonstration of An Alternative Statistic to Cohen’s Kappa for Measuring the Extent and Reliability of Agreement between Observers. Published online 2002.
Landis JR, Koch GG. The Measurement of Observer Agreement for Categorical Data. Biometrics. 1977;33(1):159. doi:10.2307/2529310
Köbel M, Piskorz AM, Lee S, et al. Optimized p53 immunohistochemistry is an accurate predictor of TP53 mutation in ovarian carcinoma. J Pathol Clin Res. 2016;2(4):247–258. doi:10.1002/CJP2.53
Jia L, Yuan Z, Wang Y, Cragun JM, Kong B, Zheng W. Primary sources of pelvic serous cancer in patients with endometrial intraepithelial carcinoma. Mod Pathol. 2015;28(1):118–127. doi:10.1038/MODPATHOL.2014.76
Köbel M, Kang EY. The Many Uses of p53 Immunohistochemistry in Gynecological Pathology: Proceedings of the ISGyP Companion Society Session at the 2020 USCAP Annual9 Meeting. Int J Gynecol Pathol. 2021;40(1):32–40. doi:10.1097/PGP.0000000000000725
Köbel M, Luo L, Grevers X, et al. Ovarian Carcinoma Histotype: Strengths and Limitations of Integrating Morphology With Immunohistochemical Predictions. Int J Gynecol Pathol. 2019;38(4):353–362. doi:10.1097/PGP.0000000000000530
Saner FAM, Takahashi K, Budden T, et al. Concurrent RB1 Loss and BRCA-Deficiency Predicts Enhanced Immunological Response and Long-Term Survival in Tubo-Ovarian High-Grade Serous Carcinoma. Clin Cancer Res. Published online 2024. doi:10.1158/1078-0432.CCR-23-3552
Kang EY, Weir A, Meagher NS, et al. CCNE1 and survival of patients with tubo-ovarian high-grade serous carcinoma: An Ovarian Tumor Tissue Analysis consortium study. Cancer. 2023;129(5):697–713. doi:10.1002/CNCR.34582
Kang EY, Millstein J, Popovic G, et al. MCM3 is a novel proliferation marker associated with longer survival for patients with tubo-ovarian high-grade serous carcinoma. Virchows Arch. 2022;480(4):855–871. doi:10.1007/S00428-021-03232-0
Le Page C, Rahimi K, Köbel M, et al. Characteristics and outcome of the COEUR Canadian validation cohort for ovarian cancer biomarkers. BMC Cancer. 2018;18(1). doi:10.1186/S12885-018-4242-8
Stanske M, Wienert S, Castillo-Tong DC, et al. Dynamics of the Intratumoral Immune Response during Progression of High-Grade Serous Ovarian Cancer. Neoplasia. 2018;20(3):280–288. doi:10.1016/J.NEO.2018.01.007
Lara OD, Krishnan S, Wang Z, et al. Tumor core biopsies adequately represent immune microenvironment of high-grade serous carcinoma. Sci Rep. 2019;9(1). doi:10.1038/S41598-019-53872-1
Rambau PF, Vierkant RA, Intermaggio MP, et al. Association of p16 expression with prognosis varies across ovarian carcinoma histotypes: an Ovarian Tumor Tissue Analysis consortium study. J Pathol Clin Res. 2018;4(4):250–261. doi:10.1002/CJP2.109
R: The R Project for Statistical Computing. Accessed February 27, 2024. https://www.r-project.org/
McHugh ML. Interrater reliability: the kappa statistic. Biochem Med (Zagreb). 2012;22(3):276. doi:10.11613/bm.2012.031
Koo TK, Li MY. A Guideline of Selecting and Reporting Intraclass Correlation Coefficients for Reliability Research. J Chiropr Med. 2016;15(2):155. doi:10.1016/J.JCM.2016.02.012
Gamer M, Lemon J, Fellows I, Singh P. irr: Various Coefficients of Interrater Reliability and Agreement. Published online 2012.
Gwet K.L. irrCAC: Computing Chance-Corrected Agreement Coefficients (CAC). (2019). Published online 2022. Accessed March 5, 2024. http://agreestat.com/
Nakagawa S, Johnson PCD, Schielzeth H. The coefficient of determination R2 and intra-class correlation coefficient from generalized linear mixed-effects models revisited and expanded. J R Soc Interface. 2017;14(134). doi:10.1098/RSIF.2017.0213

No competing interests reported.

HGSCITHSupplementarySR.docx

Download PDF

Reviewers agreed at journal
11 Sep, 2024
Reviewers agreed at journal
09 Sep, 2024
Reviewers agreed at journal
29 Jul, 2024
Reviewers invited by journal
29 Jul, 2024
Editor assigned by journal
25 Jul, 2024
Editor invited by journal
16 Jul, 2024
Submission checks completed at journal
16 Jul, 2024
First submitted to journal
11 Jul, 2024

You are reading this latest preprint version

Quantifying Intratumoral Biomarker Heterogeneity in Tubo-ovarian High-grade Serous Carcinoma to Optimize Clinical Translation

Status:

Version 1

Abstract

Figures

Introduction

Results

Evaluating ITH

Diagnostic Biomarkers

Prognostic Biomarkers

Designing validation studies for ITH

Discussion

Methods

Study Population

TMA preparation and study design

A priori and post-hoc power calculation

Statistical analysis

Declarations

Additional Information

Author Contribution

Acknowledgement

Data Availability

References

Additional Declarations

Supplementary Files

Status:

Version 1