Neoplasia has been proposed as a model of intracellular metabolic shift (1), where cells adopt an anaerobic glycolysis-dependent metabolism to sustain their uncontrolled proliferation, a phenomenon known as the Warburg effect (2). Molecular advances in the last years have revealed the extreme complexity of human tumors (3), and that intratumor heterogeneity (ITH) is a reflection of this complexity. ITH is obtaining an extraordinary importance in modern oncology during the last years (4, 5). Characteristically, ITH follows a stochastic spatial distribution along different tumor regions (6) making every single case really unique and unrepeatable. However, ITH is not a time-related tumor acquisition. In fact, a recent work has demonstrated in renal cancer that ITH appeared at very early stages in tumor development and therefore was not the result of tumor progression (7). Because ITH is high for most of the malignant tumors, it represents right now a major obstacle for the success of modern targeted therapies (8).

Many big tumors are in practice far away for being totally analyzed. In these situations, pathologists (the medical specialists handling surgical specimens and performing tumor samplings) partly sample such big tumors under the assumption that the selected samples are good representatives of the whole neoplasm. In other words, pathologists decide which parts of the tumors will be selected for histological and molecular analysis and which parts will not (the latter ones usually correspond with the tumor majority). For such a purpose, classic handling and routine sampling protocols (RP) are available for different tumor topographies in referential textbooks (9). However, these protocols were conceived in a dogmatic fashion during a period in which ITH was not a concerning issue and, surprisingly, none of them have been adapted to current necessities so far.

Recent literature shows many examples of controversial results (sometimes contradictory) in molecular analyses of the same tumor types and, even more important, deep disagreements in the performance of some expensive therapies. In this specific scenario, basic researchers, oncologists, and pathologists should first wonder how many of these apparent inconsistencies might be simply due to incomplete tumor samplings.

Representativeness of the tumor sample is an important issue in modern oncology, particularly in those cases in which therapeutic decisions are made with small core biopsies. In these cases, liquid biopsy appears in the next horizon as a promising alternative in detecting tumor heterogeneity and in identifying mutations associated with tumor aggressiveness. Furthermore, liquid biopsy could be regarded as a complementary sample to surgical specimens for the application of massive sequencing tools.

An efficient sampling of a large tumor must ensure properly ITH detection without incurring extra costs, and such a method has not yet been established in modern pathology. We have recently suggested a novel method called multisite tumor sampling (MSTS) (10) that makes use of the same number of cassettes and obtain a much better performance in detecting ITH as compared to RP without extra costs. Importantly, MSTS was clinically validated for being more efficient in detecting ITH as compared to RP on a series of clear cell renal cell carcinomas (CCRCCs) (11).

Multisite tumor sampling is based on the divide-and-conquer (DAC) algorithm (12). DAC is a useful strategy to solve complex problems in basic science (13) and has been also applied to answer biological problems (14, 15). In short, the DAC strategy consists of recursively breaking down a problem into smaller parts (divide) until these are simple enough to be solved directly (conquer). Then, partial solutions are merged to solve the original problem.

Multisite tumor sampling was shown to be an efficient method to sample large tumors for neoplasms growing in solid organs, i.e., kidney (10). However, the same protocol can be applied to any other large tumors, like those arising in soft tissues, liver, lung, testis, thyroid, breast, and others.

Some carcinomas, however, do grow with different shapes in the mucosa (excavated, plaque-like, plateau-like, polypoid), not like simple spheroids, and their growth is different depending on the axis considered, i.e., vertical or horizontal. This tumor category is represented, for instance, by carcinomas along the digestive tract and in the urinary bladder.

Gastric adenocarcinomas (GACs), colorectal adenocarcinomas (CRCs), and urothelial cell carcinomas (UCC) of the urinary bladder are substantially different neoplasms developing in similar topographic sites, since the digestive tract and the urinary bladder are hollow viscera with similar walls that include mucosa, submucosa, muscularis propria, and serosa. A significant subset of UCC and many GACs and CRCs are plaque-like shaped neoplasms having two main components of growth: superficial (radial spread) and vertical (deep invasion). Due to the specific characteristics of the local environment, it is likely that a substantial part of the relevant information for these neoplasms such as biological aggressiveness and prognosis are located in the invasion front (16–18). However, current RP for tumor sampling are not emphasizing enough the necessity of analyzing thoroughly this specific zone (9).

Pathologists around the world every day face the same dilemma when making the sampling of large tumors: where to sample and when to stop sampling. Many tumors that are totally homogeneous by the naked eye might present in fact high levels of molecular ITH, what increases the pressure to pathologists when performing the sampling. Although there exist some classic rules in relation with tumor/non-tumor interface, necrosis, hemorrhage, tumor edges, etcetera, however, focusing on the tumor itself, pathologists are sampling tumors in a quasi-blind fashion following dogmatic rules or local customs and habits, and this situation is no longer acceptable.

The solution for ITH detection must be affordable and workable at the same time, thereby balancing scientific accuracy and cost. A total tumor sampling, when possible, is the ideal solution. However, many tumors are too large and cannot be totally sampled without collapsing the laboratory workflow. In these cases, pathologists must decide how large the sampling must be. Although (as far as we know) there is no official answer to this question at this time, we have recently proposed a method to increase the number of samples in CCRCC while keeping the same cost fixed (42). We showed [in silico analysis (10), method implementation (43), and clinical validation (11)] that a better alternative for tumor sampling is possible: MSTS.

Here, we have adapted our method to the topographic peculiarities of stomach, large bowel, and urinary bladder, respectively, and have performed a similar in silico approach to the problem with colon cancer as an example. On the one hand, the modeling takes into account both microsatellite stable and unstable tumor categories in CRC. On the other hand, different densities of tumor budding, extramural venous invasion, and perineural invasion at the tumor front of invasion have also been considered in the analysis.

The goal of MSTS is the recruitment of many tumor samples for analysis, but the procedure must be adapted to specific tumor shape. Excavated and plaque-like neoplasms can be sampled with full-thickness bars of tumor tissue including the lower invasion front. Next, six to eight of these bars are placed in the same cassette for study (Figure 1). A thorough analysis of the invasion front into the wall seems mandatory in these tumors, not only for assessing ITH (histological or molecular) at any level in the vertical axis but also to discover tumor budding, and intravascular/perineural invasions, or any other morphologic feature with prognostic implications. Thick plateau-like and polypoid tumors can be sampled with a mixture of bars at the tumor invasion front and cubes at the rest, in a similar way we have previously proposed in CCRCC (10, 11). This way, MSTS is adapted to any possible situation, and many more tumor regions can be sampled without increasing laboratory costs (pathology laboratory costs are calculated by the number of paraffin blocks used for the analysis, where the higher the number of blocks, the higher the cost).

Since obtaining a large amount of these full-thickness tissue bars along several tumor slices may be perceived as too laborious by pathologists, we propose here the use of a multi-wheel rolling pasta cutter (or any similar device) (Figure 1) to get several bars in one step. Similarly, we proposed very recently a potato cutter grid for MSTS implementation in CCRCC (43).

Our accumulated experience with MSTS points to this method as the most affordable sampling strategy in large tumors to guarantee a proper ITH detection without increasing cost. It seems, however, that MSTS must be adapted to the tumor shape, on the one hand, and to the tumor topography, on the other hand. Tumors growing as 3D spheroids like those arising in the kidney, liver, lung, breast, and soft tissues, for example, are not confined by anatomical barriers and are well sampled with tissue cubes, as recently shown (11). However, tumors with more or less flattened shapes, arising in hollow viscera, are confined by well-defined anatomical barriers, namely, muscularis propria. In this context, MSTS must take into account the tumor orientation for sampling to get information at different levels of the wall. Obtaining tumor bars including (if possible) the whole tumor thickness is the best option for these cases.

We extended here the modeling approach we performed in our previous paper (10), but now the tumor shape is assumed to be a rectangle of dimensions 3 × L. So, the ITH modeling is similar to the previous one (10) and is represented by the matrix (γ)_ij ≡ {0,1} with i = 1, …, 3 (indicating rows) and j = 1, …, L (indicating column). The 0 value models homogeneity, while the 1 value models the presence of ITH at tumor position (i,j). Tumor budding (extendible to extramural venous invasion or perineural invasion) is represented by a different matrix (θ)_ij ≡ {0,1}, where the 0 and 1 values represent, respectively, absence or presence of budding in position (i,j).

Similar to our previous approach (10), two classes of ITH were modeled, random ITH (ranITH) and regional ITH (regITH). For both ranITH and regITH situations, ITH was simulated using an iterative method with h = 1, …, H steps and where the initial condition was for all the cases γ = 0. At each step h, a 2D position (i,j) is chosen at random and a value of (γ)_ij = 1 is assigned. After the H steps, the γ matrix is fixed and defines the tumor ITH configuration. The percentage of ITH density associated with a given tumor is defined as ρ≡H3×L×100 (c.f., x-axis in Figure 2).

For the ranITH situation, the matrix elements of γ were randomly generated at position (i,j), with no constraints for i and j, and this occurred for all the H steps. By contrast, for regITH, only at the first iteration (h = 1), the value of γ was assigned at position (i,j) with no constraints for i and j (using the same procedure as for ranITH), but for the following iterations (h ≥ 2), either the new chosen i or the new j was constrained to be necessarily a neighbor index of any of all the previously chosen i or j.

Budding was present only in tumor row of i = 1 (modeling the tumor invasion front). Budding was simulated using an iterative method with b = 1, …, B steps and where the initial condition was for all the cases θ = 0. At each step b, a random position j = 1, …, L was chosen and a value of (θ)_1j = 1 was assigned. After the B steps, the θ matrix is fixed and defines the configuration of tumor budding. The percentage of budding density associated with a given tumor is defined as σ≡BL×100.

For a given tumor, and after introducing fixed ITH and budding configurations defined by γ and θ, we repeated N times (and separately) two different strategies: routine protocol (RP, the one accepted in routine pathology) and our alternative, the DAC strategy. For each repetition and strategy, we calculated the number of successfully detected ITH and budding sites. Results were averaged across the N repetitions and also across M different tumors (each one with different ITH and budding configurations) (Figure 2).

Multisite tumor sampling is the best method for tumor sampling to detect ITH balancing high performance and sustainable cost. MSTS must be adapted to tumor shape and tumor location for an optimal performance.

JL and GP exposed the problem and provided an affordable solution; JC implemented the modeling approach; JL, GP, and JC wrote the final version of the manuscript and agreed with this submission.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.